A REPORT ON

STOCK PREDICTION USING MACHINE

LEARNING

SUBMITTED TO THE SAVITRIBAI PHULE PUNE UNIVERSITY,

PUNE IN THE PARTIAL FULFILLMENT OF THE REQUIREMENT FOR

OF

SEMINAR AND TECHNICAL COMMUNICATION

(THIRD YEAR ENGINEERING)

SUBMITTED BY

PRANAY SINHA 72143146B

DEPARTMENT OF COMPUTER ENGINEERING

ARMY INSTITUTE OF TECHNOLOGY

DIGHI HILLS, ALANDI ROAD, PUNE 411015

SAVITRIBAI PHULE PUNE UNIVERSITY

2022 - 2023

1
 CERTIFICATE

This is to certify that the seminar report entitles

“STOCK PREDICTION USING MACHINE LEARNING”

Submitted by

PRANAY SINHA 72143146B

is a bonafide student of this institute and the work has been carried out by him under the
supervision of Prof. Sushama Shirke and it is approved for the partial fulfillment of the
requirement of, Third-year course Seminar and Technical Communication of Savitribai Phule Pune
University.

(Prof. Sushama Shirke) (Prof. Dr. S. Dhore)

Guide HOD
Department of Computer Engineering Department of Computer Engineering

Place: Pune
Date:

2
 ACKNOWLEDGEMENT

I am overwhelmed in all humbleness and gratefulness to acknowledge my depth to all

those who have helped us to put these ideas, well above the level of simplicity and helped
us develop this into something concrete. The outcome of this project required a lot of
guidance and assistance from many people and I am extremely grateful and privileged to
be provided with it through all the helping entities.

I owe my deep gratefulness to my project guide Prof. Sushama Shirke, who gave his
attention and took an interest on our project work and steered me in the right direction
all along, throughout my project work by providing all the necessary information for
developing a good system. I am extremely thankful to him for providing his fascinating
support and guidance, despite his busy schedule.

I am thankful and fortunate enough to get constant encouragement, support and

guidance from all the teaching staff of the Computer Engineering Department which
helped me in successfully completing my project work. Also, I would like to extend
my sincere esteems to all my seniors for their timely and knowledgeable support
without which completion of this project would’ve been much harder to achieve.

PRANAY SINHA

3
 TABLE OF CONTENTS

CHAPTER TITLE PAGE NO.

Sr. No. Title of Chapter Page No.

01 Introduction 1-1
1.1 Overview 1
1.2 Motivation 1
1.3 Proposed system 1
02 Literature Survey 2-4
03 Methodology 4-8
04 Implementation 9-13
06 Results 14-15
07 Conclusion 16
References

4
List of Figures

1
 INTRODUCTION

1.1 OVERVIEW

Machine learning has various applications and one of them is to predict stock price. Stock
market is one of the most important sectors of a country's economy. Prediction of stock prices is
not easy since it is not stationary in nature . Stock prices are considered to be a very dynamic and
susceptible to quick changes because of underlying nature of the financial domain and in part
because of the mix of a known parameters (Previous day’s closing price, P/E ratio etc.) and the
unknown factors (like Election Results, Rumors etc.)
A new approach aimed to predict the stock market using machine learning is Python.
In this Machine Learning (ML) Based approach , that will be trained from the available
stocks data and gain intelligence and then uses the acquired knowledge for an accurate
prediction. In this context this study uses a machine learning technique called Support Vector
Machine (SVM) to predict stock prices
we observed out that strategies like random forest, Support vector machine and regression
algorithm. Support vector regression is a beneficial and effective gadget gaining knowledge of
approach to apprehend sample of time collection dataset we observed out that strategies like
random forest, Support vector machine and regression algorithm. Support vector regression is
a beneficial and effective gadget gaining knowledge of approach to apprehend sample of time
collection dataset we observed out that strategies like random forest, Support vector machine
and regression algorithm. Support vector regression is a beneficial and effective gadget
gaining knowledge of approach to apprehend sample of time collection dataset

1.2 MOTIVATION
The primary driver behind stock price predictions is the expected price of a stock in the future.
Environmental science, finance, and economics motivation may be helpful in a variety of
situations, including business and industry. It is possible to predict the stock's future worth. If the
prediction is more accurate then it will be very much profitable for us (some one who can
predict the stock price earlier ) .

1.3 PROPOSED SYSTEM

Propose a machine learning-based normalization for stock price prediction. Into this
system two architecture are used Regression-based Model and LSTM. The Regression-based

2
Model is employed for predicting unbroken values through some given autonomous values
Regression uses a given linear function for predicting continuous values of the most important
amongst them and made the predictions using these. LSTM architecture is able to identify the
changes in trends which show evident from the result. LSTM is identified as the best model for
the proposed methodology. This shows that the proposed system is capable of identifying some
interrelation within the data. In the stock market, there may not always follow the same cycle or
may not always be in a regular pattern for the changes that are occurred. The period of the
existence will differ and the existence of the trend is based on the companies and the sectors. For
investors, this type of analysis of trends and cycles will obtain more profit. We must use networks
like LSTM as they rely on the current information to analyse various information.

3
 1. LITERARURE SURVEY

In [1] author describe study to demonstrates how machine learning algorithms may be used to
forecast stock prices. The major goal of this research is to evaluate the efficacy of machine learning
methods with traditional statistical approaches. Another thing to think about is developing a better
technique that can anticipate stock prices and other prediction errors more accurately. The report
states that for prediction, both conventional statistical approaches and machine learning techniques
have been applied. Data splitting (cutting the data into training and test data), scaling, and feature
selection are the first steps in the process. For statistical prediction, model training is not required.
This strategy anticipates data by using the statistical mean of prior closing prices. We employ some
of these strategies. When applying machine learning techniques, the models are first used to make
predictions on test data after being trained using training data. Machine learning is done using nine
different ways. A comparison of statistical and machine learning approaches has been done in
terms of prediction performance and accuracy. It is shown that machine learning methods—in
particular, MLP and LSTM—are the most successful in forecasting stock prices for low MSE and
MAPE values after analysing each approach independently.
In [2] Author describe how trading is one of the most important operations in the financial
industry. Stock market prediction is the process of attempting to predict the future value of a stock
or other financial instrument traded on a financial exchange. This paper explains how machine
learning can be used to predict a stock. When forecasting stock prices, the majority of stockbrokers
use technical, fundamental, or time series research. Machine learning stock market forecasts are
made using the Python programming language. In this article, the author proposes a Machine
Learning (ML) approach that will learn from the stock market data that is currently available,
develop intelligence, and then use the learnt knowledge to provide an accurate prediction. In order
to anticipate stock prices for the big and small capitalizations and in the three different markets,
this study uses prices with both daily and instantaneous frequencies using a machine learning
approach called Support Vector Machine (SVM). In the study, they proposed leveraging data
obtained from multiple worldwide financial markets and machine learning algorithms to estimate
changes in stock indexes. A big collection of data collected from several worldwide financial
markets serves as the basis for the SVM algorithm's operations. Additionally, overfitting is not a
problem with SVM. Numerous machine learning-based methods are offered for forecasting the
daily movement of Market stocks. Calculation results indicate a good efficiency. The practical
trading models were built on top of their expert predictor. The model earns more money as
compared to the selected benchmarks.
In [3] There is Making precise market return estimates is quite difficult due to how unpredictable
and non-linear the financial stock markets are. Programmable techniques of prediction are now
more accurate in predicting stock prices thanks to developments in artificial intelligence and other
computing fields. In this study, artificial neural networks and random forests are used to forecast
the closing prices of five enterprises from distinct industrial sectors. The open, high, low, and close
prices of the stock are utilised to generate new variables that are inputs into the model. The models
are assessed using two popular strategic metrics, RMSE and MAPE. The low levels of these two
indicators show how well the models predict stock closing prices. Predicting stock market returns
is a challenging task since stock values vary often and depend on several variables that produce
complex patterns. The historical dataset provided accessible on the company's website only
contains a few inadequate parameters, such as high, low, open, close, adjacent close value of stock

4
prices, number of shares traded, etc. To improve the accuracy of the anticipated price value,
additional variables are built using the pre-existing variables. ANN and RF are both utilised to
make predictions about the stock's closing price the next day. It is quite clear from the comparison
study based on RMSE, MAPE, and MBE values that ANN offers superior stock price prediction
versus RF. The findings show that RMSE (0.42), MAPE (0.77), and MBE are the best values
obtained by the ANN model (0.013). Deep learning models that use financial news stories as well
as financial features such a closing price, traded volume, profit and loss statements, etc., might be
built for future work in order to obtain potentially superior results.
In [4] Author Describe Market systems are beyond anyone's capabilities for prediction due to stock
complexity. However, investors must correctly predict stock market values in order to earn a
substantial return. The ultimate objective of this study is to forecast behaviour on the Bombay
Stock Exchange (BSE). They have taken into account factors like commodity prices (crude oil,
gold, and silver), market history, and foreign exchange rate (FEX), which impact stock movement,
in order to predict how the Bombay Stock Exchange would act (BSE). The outcomes of the models
are then compared to other benchmarks. A systematic link between the different qualities used was
also constructed. It was found that there was a significant positive correlation between market
performance and gold price. The AdaBoost algorithm outperformed other approaches when
compared. The results of the study imply that machine learning algorithms may be used to predict
whether the stock market's performance would increase or decrease. It illustrates how reliant BSE
is on the factors taken into account in the study. Their findings show that BSE is most reliant on the
gold rate due to the biggest correlation relationship. Additionally, silver rate has the lowest
correlation coefficient, which suggests that BSE is least dependent on it. AdaBoost, the most
accurate machine learning technique, obtains an accuracy rate of 76.79% with 70% training data
and 75% accuracy with untrained data. There is still space for development with this project. The
project might be developed to include other topics, such interest rates, political changes, and
economic developments.
In [5] Stock price forecasting is one of the most researched topics because it appeals to both the
academic and commercial communities. Since the invention of artificial intelligence, several
algorithms have been employed to predict the movement of the stock market. For understanding
long-term markets or for predicting the opening price of the stock the very next day, machine
learning and statistics have been integrated. The many techniques for predicting share prices are
examined in this paper, including traditional machine learning, deep learning, neural networks, and
graph-based approaches. It offers an in-depth analysis of the techniques used to predict stock prices
and examines the challenges encountered as well as the direction of next studies in the area. As a
result of the stock market's close relationship to a country's economic growth, the market draws
sizeable investments from investors, and it issues securities in the public interest, forecasting the
movement of the stock prices and the market becomes essential in order to prevent significant
losses and make wise decisions. In this study, it provided a comparative comparison of multiple
algorithms for forecasting the values of different stocks. Convolutional neural networks, artificial
neural networks, long short-term memory, and other neural network models were included to the
study. The study also combines a variety of additional approaches, such as sentiment analysis, time
series analysis, and graph-based algorithms, and analyses the results of these algorithms in order to
forecast the stock prices of various corporations.

5
In [6] discussed about ML Model ,The stock market, commonly known as the stock exchange, is
one of the most involved and challenging ways to do business. Although the stock market is
unpredictable, it is one of the finest methods to make big returns when approached systematically.
Small businesses, brokerage companies, and the financial sector all need this organisation for the
generating of income and risk distribution. This work intends to use open-source tools and pre-
existing methodologies to construct machine learning models in a Their app to anticipate future
stock values for exchange, making this risky sort of transaction a little bit more predictable. The
traditional technique is circumvented, and a solution that integrates text-based machine learning
models and pattern recognition is developed to prevent having the conclusion just rely on
numerical results. The objective is to create a platform using the existing stock prediction models
that forecasts future stock market values while taking into consideration elements like news
articles, stock volume, the most recent close, etc. To predict changes in stock price, the Stock Clue
platform employs a range of ML models, including text- and series-based ones. The stock
prediction accuracy of machine learning models is substantially sufficient for practical
implementation. The news vector data, which indicates stock volatility, enables users to make
informed investments. The XGBoost model's outcomes are more accurate since it integrates time
series and news data. The current accuracy of machine learning stock predictions is over 90%.
Because most of the systems are free sources and don't need a lot of time to grasp a stock, small-
scale traders may utilise this instrument for a minimal cost.
In [7] Predicting fluctuations in the stock market is a frequent topic of attention. Today's social
media is a great mirror of the attitude and opinion of the general public about current events. For
gauging public mood, researchers have concentrated a lot of their attention on Twitter in particular.
A fascinating research was conducted on the subject of stock market forecasting utilising sentiment
analysis from Twitter. A classifier was described in earlier parts. Characteristics of it 80 percent of
the entire data is utilised to train the model, while the remaining 20 percent is used for testing. They
are the 3-day aggregate sentiment values, and their output was the rise or fall in stock price,
denoted by 1/0. The accuracy value obtained using the logistic regression method to train the
classifier was 69.01%, and it varied based on the training set. 90% of the information. The LibSVM
model they used to train achieved a result of 71.82%. These findings provide investors a significant
edge and show a direct correlation between stock market fluctuations and Twitter sentiment. This
pattern demonstrates that the models continue to work effectively as the dataset grows. In this
essay, the authors show how closely a company's stock price movements and the opinions
expressed on Twitter about it by the general public are related. The main accomplishment of their
study is the development of a sentiment analyzer that can identify the type of sentiment present in a
Tweet. The Tweets are divided into three categories: neutral, negative, and positive. It might be
argued at first that positive thoughts or feelings expressed by the public on Twitter would affect a
company's stock price.
In [8] author presented a finance and economics, stock market forecasting in a crucial time-series
learning issue. The efficient market hypothesis (EMH) claims that markets are efficient and stock
prices already accurately represent all information that is generally available, including financial
news and pricing data. This hypothesis makes use of their-form EMH, which contends that stock
prices represent a subset of public data that include all market information. Semi-strong form EMH
makes the assumption that new information is promptly absorbed by the market and that stock
prices accurately represent all newly available information. An investor theoretically cannot profit

6
or surpass the market by trading on fresh information given this supposition and the reality that
investors buy equities after publicly available information is made available. Numerous studies
debunk this claim, and through their research The main goal of this study is to extract information
about the possibility that the price of a particular stock will close higher or lower than when it
opened by using a variety of time-series and NLP approaches. The temporal nature of stock price
information and financial news is advantageous to it. In Section 2, they look at the most recent
developments in stock movement categorization and the use of deep learning to solve issues with
financial prediction. Index prediction as a whole is the subject of recent advancements in sentiment
analysis for stock markets. Its experiments, which are based on cutting-edge models, contribute to
the body of information previously accessible on the issue and show that maybe more research (in
network architecture, data size, hyper-parameter tunings, and setup) is needed to get greater
accuracies.
In [9] Author show to get the output using simple vector machine ,Investment methods for the
stock market are intricate and dependent on the analysis of enormous volumes of data. In an effort
to determine if machine learning techniques may enhance market projections over more
conventional methods, academics have recently begun to pay greater attention to them. Relevant
peer-reviewed journal papers over the past 20 years are found using a systematic literature review
strategy, and research with similar methodologies and settings are categorised.  These studies fall
within the areas of support vector machines, genetic algorithms mixed with other techniques,
hybrid or other artificial intelligence approaches, and artificial neural networks. Studies from each
category are examined to look for patterns, strange outcomes, restrictions, and areas that require
more research. A summary and recommendations for more research are included at the section's
conclusion. Support vector machines carry out they'll while assessing if the overall stock market
index is anticipated to increase or decrease. Genetic algorithms use an evolutionary problem-
solving approach to choose which stocks to include in a portfolio or to find higher quality system
inputs. Even though each research demonstrated a successful application of the methods, single
approach applications do have significant disadvantages. One approach to get around some of these
issues is to employ hybrid machine learning methods. The problem is that ultimately the systems
become unworkable due to their excessive complexity.
In [10] Author use stock certain unobserved traits to predict the data. In the era of big data, deep
learning has been much more often used to forecast stock market prices and trends. They acquired
two years' worth of data from the Chinese stock market in order to anticipate the price trend of
stock markets and provided a full customization of feature engineering and deep learning-based
models. The proposed solution is full and consists of pre-processing the stock market dataset,
several feature engineering techniques, and a customised deep learning-based system for stock
market price trend prediction.   The most recent work also recommends an analogous hybrid neural
network design, fusing a convolutional neural network with a bidirectional long short-term
memory, in order to predict the stock market index. Researchers frequently proposed different
neural network topologies as solutions, but this led to more discussions over whether the high cost
of training such models was worthwhile given the results. Their recommended solution has
advantages that are obvious. The approach is complex and recommends the best investment parts
using two distinct algorithms and data pre-processing. The system also has a forecasting
component that preserves the properties of the time series. However, there is a mistake in the
assessment area of their work. Instead of using a large dataset, they decided to test the
recommended approach on 25 recognisable stocks. There is a strong possibility that the well-
known stocks may share certain unobserved traits.

7
In [11] Author use various Machine learring algorithm to predict the upcoming data. Machine
learning (ML) is now a potent analytical tool for managing investments effectively in the financial
markets. A novel method that can assist investors in making better investment and management
decisions to achieve improved performance of their securities investments has been made possible
by the widespread usage of ML in the financial sector. Because they are appealing and relatively
liquid assets that may be sold and bought on stock exchanges, equity securities are among the most
traded securities. Due to the volatility and uncertainty of the stock market, equity investments carry
a significant risk even though they offer an appealing return. Many investors worldwide have
expressed interest in stock investing. Making a decision, however, is a challenging and complex
undertaking since there are many variables at play. Investors are eager to predict the stock market's
future condition in order to make profitable investments. Gains from even little increases in
forecast accuracy can be substantial. Because of this, stock price forecasting is a crucial procedure
that may be useful to investors. The state-of-the-art machine learning (ML) algorithms and
approaches that have been applied in finance, particularly the stock price prediction, were reviewed
and compared in this work. Numerous ML algorithms and strategies have been reviewed in terms
of input types, goals, benefits, and drawbacks. Some ML algorithms and methodologies have been
widely used for stock price prediction based on their traits, accuracy, and mistake rate.

In [12] The main objective of the presented technique is to identify the best model for predicting
stock market value. The research presents a gradient-based back propagation neural network
approach to improve optimization in stock price forecasts. Back propagation neural network
approach aims to compute the learning rate and training cycle parameters in an adjustable manner
to obtain the optimal value during the stock data training process and to generate precise forecasts.
The more realistic way to forecasting stock movement will be covered in this essay. Recent
research offer particular strategies to address these problems. The Support Vector Machine (SVM)
is a rather complex learning approach that uses the kernel method, manages the decision function,
and generates sparse solutions. In this research, they offer a theoretical and practical foundation for
using the Support Vector Machines method to anticipate the stock market. Six macroeconomic
factors that may have an impact on the stock trend are chosen for further stock multivariate
research. These elements are company-specific. Support Vector Machine is employed to evaluate
the relationship between these factors and predicted stock performance. The results show that SVM
is an effective stock forecasting tool in the financial industry.
In [13] we use various python library to predict the market, Because of the Indian stock market's
extreme unpredictability and indeterminism, which have an endless number of components that
control the directions and trends of the stock market, predicting the upswing and downturn is a
challenging process. This project will demonstrate the application of recurrent neural networks in
finance to forecast the closing price of a stock and measure market sentiment in real-time. By
combining the two approaches, the suggested model can offer buy or sell recommendations. The
recommended system has been implemented in an app made with Django and React. The React

8
App uses scraping to display the live pricing and headlines from the self-built Django Server. The
Django server also acts as a link between the React frontend and the machine learning algorithm
built with Keras and enhanced with Tensorflows. Two years' worth of historical data are scaled
using "MinMax Scaling" and then overlapped on the premise that seven days of data predict the
eighth day. The model is then trained using a two-layer LSTM model that has 64 units on the first
layer and 64 units on the second layer. The model is then collapsed into a Dense Layer with 16
input nodes and 1 output node. The "mean squared error" is used to calculate the loss, and the
Adam optimizer is then used to reduce it. 300 epochs of training have been completed with a 32-
batch size. The model is then used to predict how the stock will close on the next trading day in
order to determine the anticipated closing price.
In [14] author presented machine learning using stock market forecasting , Investors have always
had a possibility to profit on the stock market, but the bulk of gains come through study of both
current and past market circumstances, followed by proactive actions. Due to the present
overheated market situation, there are numerous elements that need to be considered before making
a productive stock market buy. It would be time-consuming and error-prone to manually analyse all
of these variables and impacting factors. Therefore, a machine learning technique will assist the
analysis of such a seemingly chaotic system the best. In this research, machine learning is being
utilised to forecast a variety of stock or index characteristics, including future opening and closing
prices as well as index values. It sought to understand the financial market's informational
environment and identify the elements that influence stock prices by taking into account the
activities taken and the extensive industrial, macroeconomic, and market data. However, this study
proved to be highly beneficial from an empirical standpoint, resulting in the invention and
implementation of a number of strategies for financial market prediction. Predicting the direction of
stock price movement is essential for the development of effective stock market operating
strategies. According to the survey's findings, 80% of eligible respondents believe their models and
methods are sufficient, but they don't utilise computer techniques to trade stocks. As a result,
employing ANN to anticipate stock behaviour and trends has shown to be an effective substitute
for conventional methodologies, giving investors access to insider information and exposing
market behaviour. It could also be applied in combination with other strategies.

In [15] Authors use Forecasting ,the future value of a company's financial stocks is the aim of stock
market prediction. A recent development in stock market prediction technology is the application
of machine learning, which generates projections based on the values of current stock market
indices by training on their past values. Machine learning itself employs a variety of models to ease
and validate prediction. The research focuses on regression and machine learning for stock value
prediction based on LSTM. Factors include open, close, low, high, and volume. In this work, a
sequential model with an output value of 256 was produced by stacking two LSTM layers. The
input to the layer is composed of two layers. A dropout value of 0.3 has been fixed, which means
that 0.3 of the total nodes will be frozen throughout the training process in order to prevent over-
fitting of the data and speed up the process.  The LSTM and regression methods have been used to
the Yahoo Finance dataset in this study. Due to an improvement in forecast accuracy, both tactics
have produced favourable results. The effectiveness of recently created machine learning strategies
for stock prediction indicates their use in profitable exchange schemes. It has led to the conclusion
that stock market forecasts may be produced more successfully and precisely by utilising machine
learning techniques. The stock market prediction system can be considerably improved in the
future by utilising a significantly larger dataset.

9
In [16] Authors combine different approaches .When attempting to anticipate the stock market, it
could be challenging to forecast future stock prices. Because equities change so much, it is very
difficult to anticipate the stock market. The value of stocks changes constantly every day. Stock
market forecasts are highly sought for by stock customers. Applying all of the extracted concepts at
once is pretty challenging in terms of making reliable stock price predictions. The newest stock
market forecasting techniques, such as Artificial Neural Networks, Neuro-Fuzzy Systems, Time
Series Linear Models (TSLM), and Recurrent Neural Networks (RNN), are investigated in this
framework study. This study's goal is to discuss several stock market forecasting techniques. One
of the stochastic techniques for applying a prediction model is the linear time series model
(TSLM). An ideal linear model is initially created using a linear time series model, and it is then
updated with data to reflect the properties of the actual data. This time series linear model's main
advantage is that it combines actual data into an ideal linear model. It is allowed to use both
traditional trends and seasonal data trends. Numerous stock market forecast parameter approaches
are collected and compared in this research. These techniques are used to evaluate stock market
performance and trends. The stock market forecasting system's accuracy should improve. This
study evaluates an unique way to enhance stock prediction outcomes by combining two or more
approaches to produce a new approach method.
In [17] The stock value is a crucial factor in maximising a company's profits. It has a direct impact
on the economy, either favourably or unfavourably. Numerous studies have been done to help
investors decide which course of action to take when buying or selling stocks. Since machine
learning and deep learning algorithms have shown to be the most reliable and efficient ways to deal
with prediction problems, they are used in the bulk of current research. Stock price prediction is a
very challenging task due to its non-linearity and wide range of sudden swings. The issue of stock
price forecasting is recognised as a classic that has garnered a lot of interest over time. This
literature review's objective is to look at the most recent deep learning and machine learning
techniques for evaluating stock price values. To maximise the use of this literature study, a
comparative analysis is conducted to quickly describe the advantages and disadvantages of each
approach. Comparisons are made to assess the efficacy and precision of various approaches.
Deep Learning algorithms are preferred due to their success in handling time series problems. The
bulk of modern techniques make use of them. Finding trends in stock price movement and applying
those trends to predict the future were the main goals of this literature review. Most sources
recommend considering business facts in order to develop accurate predictions. They advocate
using LSTM methods since they outperform the others because to their larger memory capacity.
Deep Reinforcement Learning is a key component of the most recent price prediction techniques.
DRL has proven to be successful in handling time series difficulties, such as stock price prediction.

In [18] Author analyse the twitter to get the sentiments of the people for stock
prediction .According tobehaviour economics, emotions have a considerable influence on people's
behaviour and decision-making. Here, they look into how the historical performance of the Dow
Jones Industrial Average and ratings of overall emotional states gleaned from enormous Twitter

10
feeds (DJIA). They analyse the language in daily Twitter feeds using two mood-tracking tools:
Opinion Finder, which contrasts positive and negative mood, and Google-Profile of Mood States
(GPOMS), which gauges mood on six different dimensions (Calm, Alert, Sure, Vital, Kind, and
Happy). They cross-validate them by comparing the ability of the generated mood time series to
ascertain the public's response to the 2008 presidential election and Thanksgiving Day. Then, a
Granger causality analysis and a self-organizing fuzzy neural network are used to examine whether
changes in DJIA closing values can be predicted by changes in public mood states as defined by
the Opinion Finder and GPOMS mood time series. According to their research, the accuracy of
DJIA predictions may be significantly improved by include some public mood characteristics while
excluding others. They note a decrease of more than 6% in the Mean Average Percentage Error and
an accuracy of 87.6% in predicting the daily up and down movements in the DJIA closing values.
They did not account for a number of important factors that would form the basis for future study
in their analysis. They begin by stating that neither the breadth of their study nor its relevance to a
particular place or group of people throughout the world are meant to be restricted. Given that
individuals from all over the world have an influence on US financial markets, this technique
would make sense. However, throughout the observation period, the majority of Twitter.com users
were American and English-speaking.

In [19] The Author use different Hybrid Model for Predicting stock price , Predicting the price
correlation of two assets over the next time periods is essential for portfolio optimization. They
predict the correlation between the stock values of two distinct stocks using LSTM recurrent neural
networks (RNN). RNNs are capable of understanding temporal dependencies. They also include
linearity and nonlinearity into the model using the ARIMA method. After the ARIMA model
eliminates linear trends in the data, the LSTM model is given the residual value. The ARIMA-
LSTM hybrid model is contrasted with other traditional predictive financial models, such as the
complete history model, constant correlation model, single-index model, and multi-group model.
The ARIMA-LSTM model's forecasting ability outscored all other financial models on a
significant scale in their empirical investigation. They employed the ARIMA-LSTM hybrid model
to first filter out linearity in the ARIMA modelling step and then predict nonlinear tendencies in the
LSTM recurrent neural network. The test results showed that the ARIMA-LSTM hybrid model
performs significantly better than other comparable financial models. Measures MSE, RMSE, and
MAE. They are employed to evaluate the model's effectiveness over a range of time horizons and
asset combinations. In their trial, the Constant Correlation model outperformed the other financial
models, although its values virtually halved. Based on this superior performance, they may infer
that the ARIMA-LSTM hybrid model has adequate predictive potential. As a result, the ARIMA-
LSTM model would be useful as a predictor of correlation coefficient for portfolio optimization. A
more accurate portfolio optimization may be achieved with a better predictor, increasing
investment returns.
In [20] The Author used Hierarchial Graph Attenetion Network . The study of the stock market has
long aroused the interest of many academic and business scholars. To accurately predict future
trends in stock values, several different methodologies have been developed. The use of graph-
structured data has gained popularity recently in the computer science research groups. Relational
data-based stock market prediction algorithms have just lately been introduced, thus they are still in
their infancy. First, there is a wide range in the accuracy of the data obtained from different types
of interactions. There hasn't been any prior study on the impact of various relation types on stock
market forecasting or on how to successfully combine data on multiple relation types in a

11
beneficial way. The node classification challenge is also comparable to earlier attempts in that
individual stock prediction was the only issue they tackled. In order to solve this, they propose a
hierarchical attention network for stock prediction (HATS) that uses relational data to forecast the
stock market. Each company's representations are updated with the data that their HATS approach
has carefully collected on various relation types. Specifically, node representations are initialised
using attributes extracted from a feature extraction module. HATS is used as a module with
initialised node representations in relational modelling. The newly updated node representations are
then transferred into a layer that is tailored to the task at hand. Similar to the graph classification
job, their method is used to anticipate changes in market indexes as individual stock prices. The
experiment's results show that performance might change depending on the relational data being
used. HATS surpassed all other methods because it has the capacity to pick information
autonomously.
In [21] Author use LSTM and random forest to predict the direction of price movement for
intraday trading ,They compare how well they would be able to forecast directional out-of-sample
movements in the stocks that make up the S&P 500 during intraday trading from January 1993 to
December 2018 using both random forests and LSTM networks as training methodologies. They
provide a multi-feature system that includes intraday returns, returns compared to closing prices,
and returns related to opening prices. They compare their trading strategy to Fischer & Krauss
(2018) and Krauss et al. (2017). On each trading day, they buy the 10 equities that have the greatest
chance of exceeding the market in terms of intraday returns and sell short the 10 stocks that have
the lowest chance, all of which have identical financial standing. Their empirical results show that,
before transaction costs, the multi-feature design provides a daily return of 0.54% for random
forests and 0.64% for LSTM networks. Hence They surpass Fischer & Krauss (2018) and Krauss et
al. (2017)'s single-feature settings, which only include daily returns relative to closing prices and
have equal daily returns of 0.41% and 0.39% for LSTM and random forests, respectively. They
point out that determining the direction of stock (or currency) movements can be approached as a
model-free or model-based reinforcement learning problem. It would be interesting to compare
these different reinforcement learning techniques with LSTM and random forests, especially when
applied to bitcoin.
In [22] author used Temporal Relational Ranking ,Stock prediction aims to anticipate a stock's
future patterns in order to help investors make good investment decisions. Systems for stock
prediction have often been built on the basis of time-series models. Deep neural networks have
recently become successful at modelling sequential data, and this success has led to deep learning
emerging as a possible alternative for market prediction. Most deep learning techniques now in use,
however, are not optimised for selecting the best stock with the highest anticipated return on
investment. In order to anticipate stock trends, they frequently frame stock prediction as a
classification issue or a regression problem (to predict stock price). Additionally, they primarily
view the stocks as being unconnected to one another. The important signal in the intricate linkages
between stocks (or companies), such when two stocks are in the same industry or when two firms
are suppliers to one another, is not taken into consideration. In this study, they introduce Relational
Stock Ranking (RSR), a novel deep learning technique for stock prediction. In two important
respects, their RSR technique outperforms previous strategies: (1) deep learning models are
specifically designed for stock ranking; and (2) time-sensitive stock relations are recorded. They
offer Temporal Graph Convolution, a novel neural network modelling element that jointly models
the temporal evolution and relation network of stocks, which makes their study novel. They use
historical data from the NYSE and NASDAQ stock exchanges to back-test their methods. Several

12
trials demonstrate the potency of their RSR strategy. This one outperforms contemporary stock
prediction methods, with average return ratios of 98% and 71% on the NYSE and NASDAQ,
respectively.
In [23] Author use the Artificial Counselor System .An original trading system that functions as a
synthetic stock investing advisor is suggested in this study. Support Vector Regression is employed
in this study to predict future stock prices (technical characteristics). In order to attain the largest
predicted return given their level of risk tolerance, the investor should allocate which budgetary
categories to the different current shares using the forecasted prices. Two other methods for
recommending the best portfolios are the Markowitz portfolio theory and fuzzy investment
counsellor. In contrast to the first technique, which is an optimization-based method that only
considers technical elements of the stock market, the second approach is based on fuzzy logic and
considers both technical and fundamental components of the market. The testing results on the
New York Stock Exchange show how effective the proposed approach is (NYSE). This study
introduces a ground-breaking end-to-end artificial investment adviser that anticipates the price time
series in a variety of stocks before advising the best percentages of the budget to invest in the
stocks for the highest return in the long run. Contrary to portfolio theory, the FIC gives us the
opportunity to include even additional variables, such news or emotive characteristics (such as the
impact of social media; Yang, Mo, and Liu, 2015). This enables us to function more like a
seasoned broker. To develop a trading system that is more complete, future study may use
additional specialised financial ideas like short selling and Elliott waves.

In [24] Author Use Stock Movement Prediction Using Historical Prices and Tweets, stock
movement prediction is a challenging topic since the market is very erratic and they build
temporally-dependent forecasts from disorganised data. In order to solve these three problems, they
provide a fresh deep generative model that combines text and price signals. Unlike discriminative
or topic models, their approach integrates recurrent, continuous latent variables for a better
handling of stochasticity and neural variational inference to solve the unsolvable posterior
inference. We also propose a hybrid objective with a temporal auxiliary to more readily capture
expected dependencies. They demonstrate the cutting-edge functionality of their proposed model
on a recent stock movement forecast dataset they have gathered. Stock movement forecasting is
essentially a time series problem. The relevance of the time dependence between movement
predictions is not yet addressed in NLP research. To demonstrate the effectiveness of deep
generative approaches for stock movement prediction using social media data, they created
StockNet, a neural network architecture for the task. They tested their model on a large, brand-new
dataset and discovered that it beats reliable baselines like the utilisation of earlier research.

In[25] Attention-Based Recurrent Neural Network for Stock Trend Prediction Using Financial
News ,Author used Stock market prediction is one of the most enticing study areas since correct
predictions of the market's future movement provide significant gains. The basis for traditional
short-term stock market projections is often the examination of historical market data, such as stock
prices, moving averages, or daily returns. However, information about the market and publicly
listed companies is frequently included in financial news. Although existing methods in the finance
literature make use of sentiment signal characteristics, they are limited since they don't take events
or the news's context into consideration. They are able to resolve this issue by employing deep
neural networks to extract useful semantic information from news content. A bidirectional LSTM is
specially used to encrypt the news content in order to collect context information, and a self-

13
attention approach is used to focus attention on the words, stories, and days that are the most
important. By proving that this method is competitive with other cutting-edge algorithms for
forecasting directional changes in both the Standard & Poor's 500 index and individual enterprises
stock price, they show the usefulness of current NLP technology breakthroughs for computational
finance.The success of deep learning methods in problems involving natural language processing
was the driving force behind this research. They proposed an Attention-based LSTM model to
predict the direction of changes in the Standard & Poor's 500 index and the stock prices of
particular companies (At-LSTM). Experimental results show that their strategy is competitive with
the newest model that incorporates knowledge graphs into the event embedding learning process.

In [26] Author Predict stock Market’s Reaction to News Sentiments ,For the purpose of planning
company operations, stock market forecasting is essential. The accuracy of the computer models
used to forecast the stock market is essential given how dependent it is on the economy and how it
may instantly cause financial loss. They collected, extracted, and looked at the stock market's
responses to news feelings for this study. A dictionary-based sentiment analysis model, a sentiment
analysis dictionary for the financial sector, and an assessment of the model for estimating the effect
of news sentiments on pharmaceutical stock prices are among their major achievements. With just
news emotions, they can predict patterns in short-term stock price movement with a directional
accuracy of 70.59%. Sentiment analysis, which is the technique of identifying the illogical feelings
expressed in the text, has been shown to be essential in a number of fields, including politics,
healthcare, product recommendations, and surveillance. When analysing investor opinions, it is
important to consider how moods and emotions are expressed in a sizable amount of social media
data. Utilizing Twitter data to predict stock prices based on public sentiment has gained popularity
recently. This is because stock values are instantly and quickly impacted by Twitter data, which is
already accessible to the general public. However, in order to fully utilise Twitter data, quality
controls must be put in place due to its widespread use by malicious users to promote or disparage
companies, services, ideas, and ideologies. This study use specialised sentiment analysis to
ascertain how news sentiments impact stock prices in the pharmaceutical sector. One of the major
contributions of this study is a lexicon for sentiment analysis. The sentiment ratings obtained from
the examination of news items are an effective predictor of stock movements when utilised to
appropriately leverage the forecasting of short-term trends.

In [27] Author use Deep learning Neural Network with candlestick chart Representation Since
Stock market prediction is still a challenging problem to resolve since there are so many factors
that influence stock market price, including business news and performance, industry performance,
investor attitude, social media emotion, and economic concerns. This study explores the prediction
of the stock market using deep neural networks and candlestick charts. The outcomes are utilised to
provide a framework for decision assistance that traders may use to provide indicators for the
future course of the stock price. Convolutional neural networks, residual networks, and visual
geometry group networks are only a few of the neural networks that are used in this research. They
created candlestick charts from historical stock market data. The information from these
candlestick charts will also be utilised to train a convolutional neural network model. With the help
of this Convolutional Neural Network model, we can analyse the patterns in candlestick charts and
predict future stock market movements. Their method's effectiveness in stock market prediction is
evaluated using integrating findings of 92.2% accuracy for the Taiwan stock market dataset and
92.1% accuracy for the Indonesian stock market dataset. In terms of sensitivity, specificity,

14
accuracy, and MCC, they found that the model using the CNN learning algorithm and long-term
trading days period offers the best performance. Convolutional neural networks have shown they
can anticipate future stock market movements and reveal hidden patterns in candlestick charts. The
algorithms' capacity to find hidden patterns is not actually enhanced by the addition of a volume
indicator to a candlestick chart.

In [28] Author use Deep Q-Network Stock Chart Image-Based Global Stock Market
Prediction ,They employed a Deep Q-Network with a Convolutional Neural Network function
approximator, which accepts stock chart pictures as input, to forecast the global stock market. Their
model often makes money in both domestic and foreign stock markets, not simply the stock market
of the country where it was taught. They assessed and trained their model entirely on the US
market during a 12-year period. The portfolios created using the program's output generally
provide returns per transaction in 31 different nations that range from 0.1 to 1.0 percent before
transaction costs. The results show that specific stock chart patterns can predict comparable future
stock price movements across a variety of foreign stock markets. The results also show that future
stock prices may be predicted even when training and testing are conducted in different countries.
The training method might be tested in both small and comparatively big, liquid marketplaces (like
the USA). This research demonstrates that artificial intelligence-based stock price forecasting
models may be used in relatively small markets even in the absence of training data (developing
nations). One alternate method for predicting future stock prices in the field of computer science is
to create artificial intelligence-based models that use machine learning methods like Neural
Network (NN). In order to recognise non-linear patterns in the raw input data, NN is utilised.
Convolutional neural network (CNN) or recurrent neural network (RNN) models are used in many
contemporary approaches, including speech recognition, image classification, and natural language
processing. However, unlike supervised learning, where a model receives precise answers, RL
agents are trained to maximise cumulative reward during the training process. Numerous studies
have employed RL strategies or NN-based techniques. These studies demonstrate that such AI-
based models are capable of efficiently capturing complex non-linear future return patterns among
a wide range of input variables. The majority of prior efforts, however, were focused on creating a
high-performance model that was optimised on a limited number of stocks or composite indexes in
a single country using a range of input variables, including price, volume, and technical and other
financial indicators. None of these works tested their models on the global stock market.

In [29] Author use Deep learing and certain forecasting technique to accurately identified the
stockd ,Both economists and computer scientists are interested in stock market forecasting since it
is a traditional but challenging subject. In order to develop an effective prediction model, both
linear and machine learning approaches have been studied during the past 20 years. Recently, deep
learning models have been offered as new directions for this field, and the pace of the research is
too rapid for anybody to keep up. This survey's purpose is to give an up-to-date assessment of
recent work on deep learning models for stock market prediction. They categorise the many data
sources, neural network designs, and frequently used assessment metrics, as well as taking
implementation and repeatability into account. Their goal is to keep interested researchers abreast
of the most recent developments while also making it easier for them to reproduce older studies
that acted as baselines. Based on the overview, they also suggest a few prospective lines of inquiry
for more study in this area. Prior to their study, there were only a few review papers that dealt with
stock market prediction and were released concurrently with the emergence and development of

15
deep learning approaches. Stock market predictions may have been one of several financial topics
included in these previous surveys, despite the fact that deep learning applications may have been
their major emphasis. Here, they discuss a few of them in chronological order and discuss their
sources of inspiration and unique points of view.

In [30] Author tries to address the problem of stock market prediction using artificial intelligence
(AI) approaches. Two major models for stock market forecasting may be created based on the
technical and fundamental research. The technical analysis technique uses regression machine
learning (ML) algorithms to forecast the stock price trend at the conclusion of a business day based
on past price data. On the other hand, in the fundamental analysis, ML algorithms are used to
categorise the public attitude based on news and social media. Technical analysis employs
historical price information from Yahoo Finance, whereas fundamental analysis analyses tweets
about the stock market to determine the impact of sentiment on the market's forecast. The results
show a median performance, indicating that it is premature to claim that AI can outperform stock
markets given existing AI capabilities. This paper makes an effort to address the stock market
forecasting issue using ML algorithms. The subcategories of technical and fundamental stock
market analysis are considered in order to do this. The efficiency of ML systems for stock market
forecasting is evaluated based on both of these criteria. To do this, supervised learning algorithms
are trained on labelled datasets, and evaluation metrics are used to judge how well ML algorithms
perform in the prediction process. The results show that the linear regression model can estimate
the closing price with a minimal error value when used in conjunction with technical analysis. In
the fundamental study, the SVM model also exhibits a 76% accuracy in predicting public mood.
These results indicate that while AI may predict stock price trends or societal perceptions of the
financial markets, its precision is insufficient. Additionally, while being able to accurately forecast
the same value for the following business day, linear regression can only properly estimate the
closing price with a range of inaccuracy. This makes this process unsuitable for making long-term
investments.
In [31] the author proposed a method of predicting stock crisis. A stock market collapse occurs
when stock prices across all main indexes fall by more than 10%. The process of predicting a stock
crisis is challenging since the stock market is more volatile than usual. Sell-offs in stock prices can
be attributed to a variety of factors, including business profits, geopolitical unrest, the financial
crisis, and pandemic conditions. Based on the Hybrid Feature Selection (HFS) method, they created
a model for predicting stock market crises. The process of identifying stock crises is challenging
since the stock market experiences greater swings. To the best of their knowledge based on the
literature, this is the first method to handle stock crisis prediction based on stock financial metrics
and stock price. They suggested the Hybrid Feature Selection (HFS) technique to get rid of features
that aren't important to stock financial metrics. The NB classifier approach is said to be able to
identify stocks with excellent fundamentals. Later, the RSI approach is used to detect stock
overprice. To locate the stock crisis spots, moving average statics are taken into consideration.
With the help of the XGBoost and DNN regression technique, the model's efficacy is measured.
MSE, MAE, and RMSE are used to assess the model's performance. Performance-wise, HFS-based
XGBoost outperforms HFS-based DNN technique. Therefore, to increase the model's accuracy in
subsequent work, several fundamental stock and technical characteristics might be used. They have
just looked at a few technical aspects of stock pricing.

In [32] the authors proposed a model using a Convolutional Neural Network function approximator

16
with Deep Q-Network, which uses stock chart pictures as input to forecast the global stock market.
Their model produces profits not just in the stock market of the nation whose data was used to train
it, but also typically in stock markets throughout the world. Building artificial intelligence-based
models that employ machine learning methods like Neural Network (NN) or Reinforcement
Learning (RL) is one of the other methodologies used in the field of computer science to forecast
future stock values. Currently, NN and RL are two of the most used machine learning techniques.
Modern techniques in several fields, including speech recognition, picture classification, and
natural language processing, rely on convolutional neural network (CNN) or recurrent neural
network (RNN) models. They primarily concentrate on identifying patterns in their work that often
result in a profit, not only in the local stock market of the nation whose data they utilized to train
their model, but also in regional and international stock markets. They employ the Deep Q-
Network (DQN) framework, which addresses the instability issue brought on by Q-usage learning's
of nonlinear function approximators. The training procedure is stabilized using the two techniques
of experience replay and parameter freezing. They do their training using the same techniques.
Their program uses the chart pictures of a single firm as input and selects either a Long, Neutral, or
Short move each day. It is rewarded favorably or unfavorably depending on what it does and how
the price of a firm changes as a result.

In [33] the author suggested a network which can forecast the stock market movements based on all
stock linkages. Industry and academics have long concentrated on studying stock market volatility,
and predicting stock trends is difficult. The implications of other information, such as news and
current events, on projections are not taken into account in the existing research, which focuses
more on how to collect historical price attributes into graph networks. The intricacy of stock
linkages is ignored while creating stock graphs in the majority of existing graph-based learning
techniques. They suggest a unique multilevel graph attention network (ML-GAT) for forecasting
stock market movements based on these. To be more precise, they initialize the node
representations obtained from each feature extraction module, update the stock nodes in the
isomorphic graph created from Wikidata using ML-GAT, selectively aggregate the information of
different relation types, aggregate the information to the node representation, and then feed the
result to a specific forecast layer for forecasting, completing the trend forecast of 423 stocks in the
S&P 500 index and 286 stocks in the t. The experimental investigation validates ML-state-of-the-
art GAT's performance in prediction tasks by contrasting 5 widely used techniques. The F1-score
and accuracy increased by 11.82% and 12.6%, respectively, in contrast to the best performance
benchmark model, while the average daily return and Sharpe rate grew by 5.06 percent and 94.81
percent, respectively. More importantly, the stock connections in their model can be understood
and are consistent with interactions in the actual world. The outcomes also show how crucial it is to
use relational data and financial news. Varying relational data aggregation techniques used by
GCN, TGC, and ML-GAT produce different prediction accuracies.

In [34] the author proposed a model based on heterogeneous based learners for forecasting stock
trends. Only homogeneous base learners, such as ANN, decision trees, SVM, and LSTM, or
heterogeneous base learners using a variety of methods can be used in a forecasting model that uses
decision fusion. The identical collection of heterogeneous base learners was employed in two
investigations. The projections of the foundational learners incorporated stock movement trends or
stock movement probability and were for binary categorization. On the other hand, for multi-class
classification, base learners' forecasts take stock turning indications, stock movement confidence,

17
stock risk level, or stock movement into account (e.g., up, neutral, and down). Future stock prices,
stock returns, and stock volatility are the three key topics covered by regression forecasts of base
learners. In the meanwhile, a forecasting model can utilize a variety of techniques to combine the
predictions of basic learners. The predictions' data type is a crucial consideration when choosing
fusion techniques. In general, the two most popular fusion methods for classification are voting and
tree-based approaches, whereas the two most popular fusion methods for regression are simple
averages and ANN. The forecasts produced by the base learners always had the same data format
for each forecasting model. The fusion of wide classes or even predictions with multiple data types
is a new problem for stock market prediction in the future, given the diversity of forecasts in the
stock market and the requirement for decision fusion. Future study may also focus on creating
ensembles with base learners using various methods, such the jump-diffusion model, generative
adversarial network (GAN), graph neural network (GNN), and capsule network. Furthermore,
decision fusion has only been used in a small number of research on sentiment aware prediction.

In [35] the author proposed a stock ranking based model. Making wise investing selections requires
greater consideration of the relative order or ranking of stocks than of the price or performance of a
single stock. By using stock related information in the prediction task, stock ranking performance
may be enhanced. They utilise a graph-based methodology to anticipate stock rankings and feed the
machine learning model data related to stock relationships. Given that the top-k stocks will
outperform the others in terms of profitability, investors may be interested in how they behave
when predicted. As a result, for every choice of k, the performance metric for stock ranking
prediction should be top-Heighted and bounded. They suggest a new metric called normalised rank
biassed overlap for top-k (NRBO-k) stocks for stock ranking prediction as existing evaluation
techniques do not have these characteristics. Compared to the best stock-based strategy, NRBO-k-
based investment strategies produce relative investment gains that are 0.281% to 4.928% greater.
They demonstrate how, in a graph-based method, the list-wise loss function may greatly enhance
the stock ranking performance. In three of their situations, it produces a higher NRBO-10 than the
sum of the point-wise and pair-wise losses. The amount of time required to train graph-based
systems for stock ranking prediction can be greatly decreased by node embedding methods like
Node2Vec. Additionally, when a sparse stock relation network is used, they enhance the prediction
performance by hyperparameter adjusting Node2Vec.Overall, they have shown that when a graph-
based technique is employed for stock ranking prediction, the list-wise loss is a better option. For
sparse graphs, Node2Vec can enhance or reach equivalent prediction performance.

In [36] the author suggested a model based on machine learning algorithms for stock selections.
For predicting long-term stock price trends, a common tool is the classical linear multi-factor stock
selection model. The linear model assumption, however, may not be fair given the stock market's
chaos, complexity, and dynamic nature, thus it is more important to build a more comprehensive
stock selection model based on various feature selection and nonlinear stock price trend prediction
techniques. In this study, several feature selection techniques are used to choose the features, and
time-sliding window cross-validation is used to define the parameters of the machine learning-
based stock price trend prediction models using data from the Chinese A-share market over an
eight-year period. The model works best when the random forest method is employed for both
feature selection and stock price trend prediction, according to the examination of several
integrated models. To verify the efficacy of the best model, a long-short portfolio is built using the
random forest method. The RF-RF model is also shown to have the highest return when it picks the

18
top 1% of the stocks, attaining a 29.51% annualised return, when alternative stock numbers are
used to develop the model. The annualised return from 2011 to 2018 for the new long-short
portfolio is 21.92%, while the maximum drawdown is just 13.58%. The stratified back-testing
approach is utilised to further analyse the profitability of the RF-RF model. As a result, the RF-RF
model may be used to help guide investing decisions since it is quite accurate in predicting long-
term stock price patterns.

In [37] the author proposed a new model for prediction, feature extraction and optimum feature
selection. The prediction for the stock price makes an effort to predict probable changes in the
stock value on the financial exchange. The accuracy of the share price movement prediction would
increase investor returns. The three main steps of the novel stock market prediction model
presented in this study are prediction, feature extraction, and optimum feature selection. The
obtained stock market data is first processed to extract statistical parameters including mean,
standard deviation, variance, skewness, and kurtosis. Additionally, typical indicators like as
Average True Range (ATR), Exponential Moving Average (EMA), Relative Strength Index (RSI),
and Rate of Change are produced using the index data received (ROC). It is more important to
choose the most pertinent characteristics in order to achieve the greatest expected results. In this
way, a novel hybrid model known as the Red Deer Adopted Wolf Algorithm selects the best
characteristics among the retrieved features (technical indicators-based features, statistical features)
(RDAWA). Additionally, the ensemble approach is used to the chosen attributes in order to
forecast stock movement. Support Vector Machine (SVM), Random Forest1 (RF1), Random
Forest2 (RF2), and optimised Neural Network (NN) are examples of classifiers used in the
ensemble approach. The optimised neural network provides the final projected outcomes (NN). The
suggested RDAWA trains the NN by fine-tuning the optimum in order to make the exact
prediction. Finally, the performance of the suggested work is evaluated in comparison to other
traditional models using certain metrics. It is recommended as a suitable method for accurate close
price prediction of the Saudi stock market.

In [38] the author proposed a neural network that predicts stock changes based on news and
affection. In time series prediction research, which is crucial for lowering investment risk, stock
price volatility forecasting is a popular issue. The price of a stock, however, is influenced by
associated societal issues as well as its historical tendency. In this study, a hybrid time-series
predictive neural network (HTPNN) is proposed that blends news and affection. The characteristics
of news headlines are expressed as distributed word vectors, which are dimensionally reduced via
sparse automated encoders to increase the model's effectiveness. The daily K-line data is then
integrated with the news, taking into account how quickly equities change. HTPNN learns the
fusion feature of news and time series, which not only maintains the useful information of news
and stock data but also gets rid of the text's superfluous information, to capture the prospective law
of stock price fluctuation. Their approach combines more plentiful stock traits and has greater
advantages in running speed when compared to state-of-the-art methods. In addition, accuracy
often increases by over 5%. They have predetermined segmentation lengths for news and index
sequences, it should be mentioned. In actual trading, however, the impact of various events on
stock movement may vary. Future studies will examine the effects of event intensity and the best
ways to split the sequence window scientifically for greater prediction accuracy. By utilising the
investment method, they will finally accomplish the aim of profitability.

19
In [39] the author suggested a method to predict risks in stock trends. Due to a variety of deciding
variables, the nature of stock market movement has long been unclear to investors. This work uses
machine learning and deep learning techniques to drastically lower the risk associated with trend
prediction. Petroleum, non-metallic minerals, and basic metals from the Tehran Stock Exchange
are chosen for experimental assessments for their stock market groupings, namely diversified
financials. In this study, two potent deep learning techniques (Recurrent Neural Network (RNN)
and Long Short-Term Memory) are compared to nine machine learning models (Decision Tree,
Random Forest, Adaptive Boosting (Adaboost), eXtreme Gradient Boosting (XGBoost), Support
Vector Classifier (SVC), Naive Bayes, K-Nearest Neighbors (KNN), Logistic Regression, and
Artificial Neural Network (ANN)) (LSTM). They are intended to be used in two different ways,
and their input values are 10 technical indicators taken from ten years of historical data. First, stock
trading values are used to calculate the indicators, and then, before use, the indicators are converted
to binary data. Based on the input methods, each prediction model is assessed using three metrics.
The assessment findings show that RNN and LSTM perform much better than other prediction
models for continuous data. Additionally, the findings demonstrate that those deep learning
techniques are the best for evaluating binary data; nevertheless, the gap between them is
diminishing as a consequence of the second method's clearly improved model performance. Their
experimental results screamed that using binary data instead of continuous data significantly
improved the performance of models. Indeed, both techniques use deep learning algorithms (RNN
and LSTM) and their better models.

In [40] the author proposed a algorithm predicting stock index movements. In the literature, there
are several methods for determining instantaneous frequency; however, due to a wide range of real-
world issues, a particular method is not always appropriate. A novel form of instantaneous
frequency suited for a broad category of signals termed simple waves is proposed in this study. The
term "counting instantaneous frequency (IF) that should have independent interest" refers to the
new instantaneous frequency. In this research, a stock index movement prediction method based on
the newly specified instantaneous frequency and signal decomposition is provided. Based on Hong
Kong Hang Seng Index stock market data, an experiment is offered. By comparing the
effectiveness of the suggested methodology to the BP neural network prediction method, it is
determined to be effective and shows promise. It is suggested to use a brand-new instantaneous
frequency termed counting IF. The three conditions for instantaneous frequency are met by the
newly defined IF. The idea is clear-cut and easy to understand. Simple waves, including IMFs
produced from an EMD algorithm, may be created with it. It is of independent importance due to
its broad application and simple conceptualization. Following the usage of EMD decomposition,
the suggested -counting IF is then employed as an essential application to analyse stock indexes.
The counting IF approach is expected to have important applications in the study of financial data.

In [41] the author suggested a model based on long-short term memory for stock ranking
prediction. Predicting stock rankings and creating suitable portfolio strategies have grown to be
popular study areas for many academics due to the current economic and financial market
globalization. However, because the stock market has many styles at various times, changing
market styles will significantly reduce the model's ability to forecast outcomes. They suggest an
adversarial game neural network model based on LSTM and an attention mechanism for stock
ranking prediction in order to reduce the impact of style exposure and improve the stock selection
performance of the deep learning model. In order to improve the network, they additionally merged

20
trading duties to create an MS-WRSE loss function that takes stock rankings into account. In
contrast to traditional time series prediction models, the adversarial game neural network uses a
mutual game between the primary neural network and the auxiliary neural network to reduce the
impact of market-style elements on stock ranking forecasts. This may improve the model's stock
selection performance. The adversarial game neural network, however, has several drawbacks. The
output results of each round will interact due to the distinction between the primary network
learning task and the auxiliary network learning task. It is challenging to guarantee that the two
networks consistently promote one another favorably. Furthermore, the learning outcomes of the
auxiliary network cannot be used to infer if there is a possible relationship between input
characteristics and target values in the auxiliary network. In order to identify distinct possible
elements as network input and make experimental results more plausible, it is required to draw on
human past knowledge.
In [42] the author constructs a multi-Source multiple instance model that can efficiently aggregate
events, attitudes, as well as the quantitative data into a complete framework in order to better
prediction for stock market composite index movements. The challenge of predicting stock market
changes is difficult and crucial. As the amount of data increases, researchers start to glean useful
indications (such as the events and feelings) from it in an effort to make predictions easier.
However, the indicators from earlier research are frequently based on only one data source and may
not entirely account for all the variables that might influence stock market movements. In this
study, the authors construct a multi-Source multiple instance model that can efficiently aggregate
events, attitudes, as well as the quantitative data into a complete framework in order to better
prediction for stock market composite index movements. They successfully implement an unique
event extraction and representation approach in order to accurately capture the news events.
Evaluations of the data from the years 2015 and 2016 show how well their model works.
Additionally, their method can automatically assess each data source's significance and pinpoint the
key input data thought to be responsible for the movements, rendering the forecasts understandable.
Their model successfully integrates heterogeneous information—that is, the events, sentiments, and
historical quantitative features—into a comprehensive framework and takes into account the
consistency among different data Sources to make better predictions, in contrast to earlier studies
that frequently exploit only one data Source. Additionally, they suggest a brand-new event
representation learning approach that can efficiently record event data. The efficiency of their
approach has been thoroughly assessed using data from the previous two years.

In [43] the author proposed a CNN model for stock prediction. In order to anticipate potential
future prices or directions of an index or stock, stock market forecasting is a time series issue. It is
difficult to accomplish the aim using conventional time series approaches since the stock data has
significant levels of uncertainty and is impacted by a wide range of variables. Convolutional neural
network (CNN) models have been successfully employed in literature for stock market forecasting.
However, when taking these models into account, data imbalance caused by labelling and feature
selection issues are evident. Therefore, to address the problems, this study suggested a new rule-
based labelling algorithm and a new feature selection method. In order to assess the efficacy of the
data labelling and feature selection strategy, a CNN-based model that was created to forecast the
trade action of stocks in the Dow30 index the next day was presented. several sets of image-based
input variables. They are produced utilising data on the price of gold and oil as well as technical
indicators to feed the CNN model. Other papers in the literature were used to compare the
prediction performance of CNN models. According to the experimental findings, the CNN

21
prediction model, which employs the feature selection and labelling strategies suggested in this
work, performs with 3-22% greater accuracy than CNN-based models used in previous studies.
Additionally, the suggested labelling method outperforms Chen and Huang's data. They used a
lighting strategy to address the issue of stock data imbalance. The ratio between labelled data was
decreased by this technique from 15 times to 1.8 times. These findings demonstrate the
applicability of the suggested methods for stock price prediction.

In [44] the authors proposed a new model that uses a cutting-edge non-linear filtering technique for
stock analysis. One of the most crucial jobs in financial decision-making is creating an accurate
estimate of volatility. The performance benefits from hybrid models, which combine artificial
neural networks with GARCH-type models, have recently been discovered to be exceptional.
However, given the way financial data is distributed, there haven't been many investigations of
hybrid models. Since the high frequency region, which is close to zero, is the only place where
Heights in the networks can be trained to obtain accurate prediction, the distribution of volatility
time-series is highly concentrated there. This aspect can lead to low prediction performance across
the entire domain of probability density function. To address the issue, they suggest a brand-new
hybrid model that incorporates GARCH-type models and uses a cutting-edge non-linear filtering
technique. They use root-type functions for the filtering, which change the original volatility's
strongly left-biased and pointed distribution into a volume-upped (VU) distribution that is moved
to the right. The primary implementation model is long short-term memory (LSTM), and the
presented models are used to forecast the realised volatility of the S&P 500. It is discovered that the
suggested hybrid model (VU-GARCH-LSTM) achieves a performance boost of 21.03% when
compared to the mean performances of the existing hybrid models combining LSTM and GARCH
type models in terms of the root mean square error (RMSE). By making the prediction distribution
similar to the label distribution, the suggested model further enhances prediction performance in
the proper domain area of label probability density.

In [45] the author proposed a model that integrates time-varying effective transfer entropy (ETE)
with multiple machine learning techniques to forecast the direction of US stock prices. By
examining the correlation between financial crises and Granger-causal linkages among the stocks,
they first investigate whether the ETE based on 3 and 6 months moving windows may be regarded
as the market explanatory variable. Then, they find that adding the ETE driven variable as a new
feature to the logistic regression, multilayer perceptron, random forest, XGBoost, and long short-
term memory network may increase the prediction performance on the direction of the stock price.
As a forecast performance metric, they advise using the modified accuracy obtained from the risk-
adjusted return in finance. Finally, they confirm that the long short-term memory network and
multilayer perceptron are better options for stock price prediction. This work is the first effort to
forecast stock price direction using ETE, which is easily applicable in the real world.

In [46] the authors provided a methodology for predicting stock prices that takes into account how
different enterprises are connected. Stocks that are fundamentally related to one another often move
in tandem. Stock movement forecasting activities are thought to profit from taking into account
these prevalent tendencies. However, because the links between stocks are not physically exhibited
and must instead be extrapolated from fluctuating data, such signals are not simple to model. In
response to this fact, they provide a methodology for predicting stock prices that takes into account
how different enterprises are connected. They also create a new pipeline in order to efficiently

22
employ a substantial number of key attributes. They first lower the dimension of stock fundamental
information using variational autoencoder (VAE), and then they group stocks into a graph structure
(fundamentally clustering). Second, a hybrid model for graph-structured stock market forecasting
that combines long-short term memory and graph convolutional networks (GCN-LSTM) with an
adjacency graph matrix (learned through VAE) is suggested. Experiments on minute-level data
from the U.S. stock market show that their model catches both spatial and temporal information
effectively and improves over baseline approaches. The suggested approach has promise for
additional situations where a potential but undetected spatial dependence might be used to enhance
time-series prediction. However, there is compelling evidence that the interactions between
enterprises influence the movement of the stock price. According to experimental findings, their
model works better on real-world minute-level stock price data than other cutting-edge approaches.
The enhancement of predicting accuracy is influenced by fundamental features expressed in a
spatial structure.

In [47] the author offered a sliding window metaheuristic optimization-based intelligent time series
prediction system. The study and modelling of financial time series play a significant role in
guiding investors' decisions and trades, and time series forecasting has been routinely utilised to
predict future stock values. Additionally, the time series' nonlinearity is prominent in a dynamic
environment like the stock market, which has an immediate impact on the accuracy of stock price
projections. In order to accurately anticipate the stock prices of Taiwan construction businesses,
this research offers a sliding window metaheuristic optimization-based intelligent time series
prediction system. Home brokers who lack the expertise to invest in such firms may find it to be of
great interest. The programme operates as a standalone application and features a graphical user
interface. The created hybrid system shown exceptional prediction performance and raises total
investment performance profit. The suggested model is an effective forecasting method for highly
nonlinear time series, whose patterns are challenging for conventional models to grasp. Due to the
difficulty of solving extensive mathematical loops in the MATLAB programme, the suggested
system's computational speed is a drawback, particularly with respect to sliding-window validation.
With more validations, the computational cost rises. The requirement to configure several system
parameters (including MetaFA and time series parameters), despite the availability of default
options, is another thickening. Furthermore, the method does not deliver exceptional returns for
long-term investment—a fact that will spur further investigation.

In [48] the author suggested a framework for predicting the movements of stock prices using
sentiment analysis and financial news. Stock price changes are largely influenced by financial
news, which has been demonstrated to be a key determinant. Prior research, however, mostly
focused on analysing superficial aspects and disregarded the structural relationship between words
in a phrase. The association between investors' responses and news events has been attempted to
identify in several sentiment analysis research. However, the linguistic dataset, which is unrelated
to the financial industry and has poor performance, was typically used to create the sentiment
dataset. In this study, a novel framework for predicting the movements of stock prices using
sentiment analysis and financial news is proposed. The two innovative two-stream gated recurrent
unit networks proposed in this paper—Stock2Vec, a sentiment word embedding trained on the
Harvard IV-4 and financial news datasets—are the paper's original contributions. There are two
main experiments carried out: the first experiment forecasts the direction of the S&P 500 index
stock price using historical S&P 500 prices and articles crawled from Reuters and Bloomberg, and

23
the second experiment forecasts the direction of the VN-index price trend using Viet Stock news
and stock prices from cophieu68. Results reveal that: 1) two-stream GRU beats cutting-edge
models; 2) Stock2Vec is more successful at handling financial datasets; and 3) when used, a
simulated scenario demonstrates the model's suitability for the stock market. Additionally, a
simulation system was used to calculate the actual earnings investors would make using their
approach, and it turned out to be resilient against market volatility and able to adapt to the risk from
the actual market. In order to assist investors in trading certain equities, this framework may also be
included into an automated system.

In [49] the authors created a deep convolutional fuzzy system which can be used for predicting
stock market changes. A deep convolutional fuzzy system (DCFS) is a multilayer connection of
multiple low-dimensional fuzzy systems, where the low-dimensional fuzzy systems' input variables
are chosen using a sliding window over the layers' input spaces. They suggest a bottom-up, layer-
by-layer design approach for the DCFS that is based on input-output data pairs. The first-layer
fuzzy systems are specifically viewed as a They create these fuzzy systems utilising the Wang-
Mendel approach, estimating the output based only on a relatively tiny subset of the input
variables. Following the creation of the first-layer fuzzy systems, the data is passed through the
first layer to create a new dataset, and based on this new dataset, the second-layer fuzzy systems
are developed in the same manner as the first-layer fuzzy systems. Using this method again and
again, they create the whole DCFS. They also suggest a DCFS with parameter sharing to conserve
computation and memory. They use the DCFS models to forecast the actual Hang Seng Index of
the Hong Kong stock market as well as a synthetic chaotic plus random time series. It may be used
with basic devices: The user-end basic devices such as a cell phone may be used to perform the
online training algorithms due to their simple computation and minimal memory demand. It
encourages online education: Users can develop a variety of user-specific intelligent systems by
updating the DCFS models regularly with their own new data using their own low-tech devices. It
is also appropriate for situations with high dimensions.

In [50] the author suggested to create an enhanced Long Short-Term Memory (LSTM) neural
network stock prediction model using a Genetic Algorithm (GA) for feature selection. A suitable
data base for their stock price projection may be found in the financial market's abundance of
indicators used to characterise changes in stock price. Due to their various industrial sectors and
geographical locations, different stocks are impacted by various variables. Finding a multi-factor
combination that is appropriate for a certain stock in order to forecast its price is therefore crucial.
First, they rate the relevance of the various elements using the GA. Then, using the trial-and-error
process, the ideal combination of criteria is found from this ranking. Finally, they combine the
LSTM model with optimum factors to forecast stocks. The China Construction Bank dataset and
the CSI 300 stock dataset, used in extensive empirical research, show that the GA-LSTM model
outperforms all baseline models for time series prediction. Although the stock price prediction
model presented in this research has significant resilience and may effectively enhance forecast
accuracy, there are still certain drawbacks as listed below: In order for future study to include data
from many stock markets, they first restrict their experiment to Chinese stock data. Second, rather
than using a systematic technique to determine the ideal size of parameters, such as the choice of
the number of factors, trial and error is typically used in this paper's design of the model
parameters. The strategy for improvement is to integrate with other machine learning technologies
in order to identify the ideal parameters and raise the model's interpretability. Additionally, a range

24
of appropriate combinations may be produced to enhance the performance of the study when the
control parameters of GA, such as crossover rate, mutation rate, and number of factor
combinations, are established.

25
3. Methodology

Fig Proposed Workflow

The system presented here composes of five modules:-

1. Input as Dataset
2. Pre processing
3. Data splitting
4. Build & Model train Lstm, CNN and Hybrid approach of LSTM+CNN
5. Output as Predicted Result

Attribute such as: price of open, high, low, close, adjusted close price taken from huge dataset
are fed as input to the models for training to pre-process the data techniques like
normalization & one hot encoding in applied on dataset. After this data is divided in two sets
namely training & testing which are ratio of 80:20 respectively. Then, this set are used to
train a model using 3 different approaches: LSTM, CNN and Hybrid approach of
LSTM+CNNS. Finally, all these modules are evaluated using Root mean square error.

26
 Working of LSTM model

Fig1 LSTM Architecture

Long Short Term Memory is a kind of recurrent neural network. In RNN output from the last
step is fed as input within the present step. It tackled the matter of long-term dependencies of
RNN within which the RNN will not predict the word hold on within the long term memory
however can offer additional accurate forecasts from the recent info. Because the gap length
will increases RNN does not offer an economical performance. LSTM will by default retain
the knowledge for a long period of time. It is used for processing, predicting and classifying
on the basis of time-series data.

Structure of LSTM:
 LSTM has a chain organization that contains four neural networks and different memory
blocks called cells.
 LSTM has a new structure called a memory cell. The memory cell makes the decisions
about what information to store, and when to allow reading, writing and forgetting.
 A memory cell contains three main gates:
o Input gate- a new value flows into the memory cell.
o Forget gate- a value remains in the memory cell.
o Output gate- value in the memory cell is used to compute the output.

27
 Applications of LSTM includes:
 Language Modelling
 Machine Translation
 Image Captioning
 Handwriting generation
 Question Answering Chatbot

28
 Working of CNN model

Fig 2 CNN Architecture

 Layer of CNN model:

o Convolution
o MAX Pooling
o Dropout
o Flatten
o Dense
o Activation
 Convolution: In the Convolution extract the featured from the input image. It given the
output in matrix form.
 MAX Pooling: In the MAX polling it takes the largest element from a rectified feature map.
 Dropout: Dropout is randomly selected neurons are ignored during training.
 Flatten: Flatten feed output into a fully connected layer. It gives data in list form.
 Dense: A Linear operation in which every input is connected to every output by weight. It
followed by a nonlinear activation function.
 Activation: It used sigmoid function and predict the probability 0 and 1.

Applications of CNN
includes:
 Decoding Facial Recognition
 Analyzing Documents

29
 Hybrid Approach of LSTM + CNN
In the hybrid approach, the Convolutional Neural Networks (CNNs) offer benefits in
choosing sensible options and Long Short-Term Memory (LSTM) networks have proven
sensible skills to find out to learn sequential data. Each approaches are reported to produce
improved result. CNNs to possess to convolute filters over every input layer so as to get the
simple options and CNNs have shown enhancements in computer vision, natural language
processing and different tasks . CNN may be a powerful tool to pick out features in order to
improve the prediction accuracy . The capabilities of LSTMs in learning data series by
considering the previous outputs .

They have numerous completely different sizes to generate different features. The Max-
pooling layer is to calculate the most value as a corresponding feature to a particular filter.
The output vectors of the Max-pooling layer become inputs to the LSTM networks to
measure the long-run dependencies of feature sequences. One in all the benefits of the
LSTMs is that the ability to capture the sequential data by considering the previous data. This
layer takes the output vectors from the dropout layer as inputs. This layer include a set number
of units or cells and also the input of every cell is that the output from the dropout layer. The
final output of this layer has the same number of units within the network the outputs from
LSTMs are merged and combined in one matrix then passed to a fully connected layer. The
array is converted into a single output in the range
between 0 and 1 using the fully connected layer, in order to be finally classified using

4. Implmentation

Step 1: Dataset Analysis

30
 Fig 3 Stock Dataset Information
Data analysis for stock price of companies. Fig. represent the date, open, close, high, low,
adjusted close and volume of stocks details.

Step 2: Read Dataset

Fig 4 Read Dataset

After performing data analysis, I have read the dataset. It shows the dataset information table
starting from the tail. There are 4274 data are available in each companies dataset.

31
Step 3: Graph of Close Price history

Fig 5 Graph of Close Price history

Step 4: Preprocessing

Fig 6 Data Scaling

After Dataset reading, performed preprocessing operation on the dataset. Min-Max Scaler to
preprocess the dataset. In preprocessing operation removes the noise into the data and convert
data into 0 to 1 form.

Step 5: Train test Split

After performing preprocessing divide the dataset into training and testing set. 80% of the
data is used for the training while the remaining 20% of the data is used for testing..

32
Step 6: Model fitting of Long Short Term Memory architecture , Convolution Neural Network
architecture & Hybride Approach of LSTM+CNN.

Fig 7 LSTM Summary

Fig 8 CNN Summary

33
 Fig 9 Hybrid Approach of LSTM + CNN Summary

After generating training dataset, to apply training , create LSTM, CNN & Hybride Approach
of LSTM + CNN network using KERAS. several variations of this architecture using various
numbers of layers and various size of Bottleneck layer.

Step 7: Apply Training

Fig 10 Training Process

To apply training, from the samples of Training data, 1543 samples are used for training
and 460 samples are used for validation. Data is processed in a batch size of 1 and epoch is
1 for the entire training dataset.

34
Step 8: Predicted Result and Then predict Graphs

Fig 11 Predicted Close Price

Sector Stock Name RMSE

LSTM CNN LSTM+CNN
Banking ICICI Bank 22.5409 8.1499 9.1497
Pharma Sun Pharma 19.4190 16.2015 16.0716
Petroleum GSFC 5.4396 6.6478 4.6335
Software RS Software 4.7152 3.7276 3.0633
Textiles Vardmn Ploy 1.3909 2.5984 2.2982

Table Accuracy

35
 RESULT
So by doing many training on dataset get the below output

Fig Plot for Real vs Predicted value for ICICI Bank using LSTM

Fig : Plot for Real vs Predicted value for ICICI Bank using CNN

36
Plot for Real vs Predicted value for ICICI Bank using LSTM + CNN

37
 Chapter 5. Conclusion And Future Work

In report, we will compare a machine learning models like LSTM model, the CNN
model and also the hybrid approach of LSTM + CNN model. We have a tendency to
train the model using the data of NSE listed companies to predict the stock future value.
This is shows the proposed method is capable to distinctive around interrelation with the
data. Also, it is evident from the results that, Hybrid approach of LSTM+CNN model is
capable to identify the changes in trends. For the projected method the Hybrid approach
of LSTM+CNN is known as the best model. It uses the information given at a specific
instant for prediction. Even if the other two models LSTM and CNN are utilized in a lot
of other time-dependent data analysis, it is not outperforming over the Hybrid approach
of LSTM+CNN architecture in this case. This is often because of quick changes occur in
stock market. The changes in the stock market is not always be in a regular pattern or not
always follow the continuous cycle. Based on the companies and sectors, the existence
of the trends and the period of their existence will differ. The analysis of this type of
cycles and trends can offer a more profit to the investors. In future work, we add more
stock market data and compare more model to improve accuracy of predicted stock
price.

38
 REFERENCES

[1] Bhattacharjee and P. Bhattacharja, "Stock Price Prediction: A Comparative Study between
Traditional Statistical Approach and Machine Learning Approach," 2019 4th International
Conference on Electrical Information and Communication Technology (EICT), 2019, pp. 1-6, doi:
10.1109/EICT48899.2019.9068850.
[2] Vanukuru, Kranthi. (2018). Stock Market Prediction Using Machine Learning.
10.13140/RG.2.2.12300.77448.
[3] Vijh, Mehar, Deeksha Chandola, Vinay Anand Tikkiwal and Arun Kumar. “Stock Closing
Price Prediction using Machine Learning Techniques.” Procedia Computer Science 167 (2020):
599-606.
[4] Kohli, Pahul & Zargar, Seerat & Arora, Shriya & Gupta, Parimal. (2019). Stock Prediction
Using Machine Learning Algorithms: SIGMA 2018, Volume 1. 10.1007/978-981-13-1819-1_38.
[5] Soni, Payal & Tewari, Yogya & Krishnan, Deepa. (2022). Machine Learning Approaches in
Stock Price Prediction: A Systematic Review. Journal of Physics: Conference Series. 2161.
012065. 10.1088/1742-6596/2161/1/012065.
[6] A M Pranav, Sujooda S, Jerin Babu, Amal Chandran, Anoop S, 2021, StockClue: Stock
Prediction using Machine Learning, INTERNATIONAL JOURNAL OF ENGINEERING
RESEARCH & TECHNOLOGY (IJERT) NCREIS – 2021 (Volume 09 – Issue 13)
[7] V. S. Pagolu, K. N. Reddy, G. Panda and B. Majhi, "Sentiment analysis of Twitter data for
predicting stock market movements," 2016 International Conference on Signal Processing,
Communication, Power and Embedded System (SCOPES), 2016, pp. 1345-1350, doi:
10.1109/SCOPES.2016.7955659.
[8] Prosky, Jordan, Xingyou Song, Andrew Tan and Michael Zhao. “Sentiment Predictability for
Stocks.” ArXiv abs/1712.05785 (2017): n. pag.
[9] Strader, Troy J.; Rozycki, John J.; ROOT, THOMAS H.; and Huang, Yu-Hsiang (John)
(2020) "Machine Learning Stock Market Prediction Studies: Review and Research
Directions," Journal of International Technology and Information Management: Vol. 28: Iss. 4,
Article 3.
[10] Shen, Jingyi & Shafiq, M.. (2020). Short-term stock market price trend prediction using a
comprehensive deep learning system. Journal of Big Data. 7. 10.1186/s40537-020-00333-6.
[11] Obthong, Mehtabhorn, Nongnuch Tantisantiwong, Watthanasak Jeamwatthanachai and
Gary B. Wills. “A Survey on Machine Learning for Stock Price Prediction: Algorithms and
Techniques.” International Conference on Finance, Economics, Management and IT
Business (2020).
39
[12] Ajinkya Rajkar , Aayush Kumaria , Aniket Raut , Nilima Kulkarni, 2021, Stock Market
Price Prediction and Analysis, INTERNATIONAL JOURNAL OF ENGINEERING RESEARCH
& TECHNOLOGY (IJERT) Volume 10, Issue 06 (June 2021),
[13] Deshmukh, Yogita & Saratkar, Deepmala & Tiwari, Yash. (2019). Stock Market Prediction
Using Machine Learning. IJARCCE. 8. 31-35. 10.17148/IJARCCE.2019.8107.
[14] Chouhan, Lokesh & Agarwal, Navanshu & Parmar, Ishita & Saxena, Sheirsh & Arora,
Ridam & Gupta, Shikhin & Dhiman, Himanshu. (2018). Stock Market Prediction Using Machine
Learning. 10.1109/ICSCCC.2018.8703332.
[15] Polamuri, Dr & Srinivas, Kudipudi & Mohan, A.. (2020). A Survey on Stock Market
Prediction Using Machine Learning Techniques. 10.1007/978-981-15-1420-3_101.
[17] Awad, Alamir Labib. “ROLE OF MACHINE LEARNING IN PREDICTING STOCK
PRICES: A LITERATURE SURVEY.” (2021).
[18] Bollen, Johan, Huina Mao and Xiao-jun Zeng. “Twitter mood predicts the stock
market.” ArXiv abs/1010.3003 (2011): n. pag.
[20] Kim, Raehyun & So, Chan & Jeong, Minbyul & Lee, Sanghoon & Kim, Jinkyu & Kang,
Jaewoo. (2019). HATS: A Hierarchical Graph Attention Network for Stock Movement Prediction.
[19] Choi, Hyeong Kyu. “Stock Price Correlation Coefficient Prediction with ARIMA-LSTM
Hybrid Model.” ArXiv abs/1808.01560 (2018): n. pag.
[21] Ghosh, Pushpendu & Neufeld, Ariel & Sahoo, Jajati. (2020). Forecasting directional
movements of stock prices for intraday trading using LSTM and random forests.
[22] Feng, Fuli & He, Xiangnan & Wang, Xiang & Luo, Cheng & Liu, Yiqun & Chua, Tat-Seng.
(2019). Temporal Relational Ranking for Stock Prediction. ACM Transactions on Information
Systems. 37. 1-30. 10.1145/3309547.
[23] Nekoeiqachkanloo, Hadi & Ghojogh, Benyamin & Pasand, Ali & Crowley, Mark. (2019).
Artificial Counselor System For Stock Investment. Proceedings of the AAAI Conference on
Artificial Intelligence. 33. 10.1609/aaai.v33i01.33019558.
[24] Xu, Yumo & Cohen, Shay. (2018). Stock Movement Prediction from Tweets and Historical
Prices. 1970-1979. 10.18653/v1/P18-1183.
[25] Liu, Huicheng. (2018). Leveraging Financial News for Stock Trend Prediction with
Attention-Based Recurrent Neural Network.
[26] Shah, Dev & Isah, Haruna & Zulkernine, Farhana. (2018). Predicting the Effects of News
Sentiments on the Stock Market. 4705-4708. 10.1109/BigData.2018.8621884.
[27] Kusuma, Rosdyana Mangir Irawan, Trang-Thi Ho, Wei-Chun Kao, Yu-Yen Ou and Kai-
Lung Hua. “Using Deep Learning Neural Networks and Candlestick Chart Representation to
Predict Stock Market.” ArXiv abs/1903.12258 (2019): n. pag.
[28] J. Lee, R. Kim, Y. Koh and J. Kang, "Global Stock Market Prediction Based on Stock Chart
Images Using Deep Q-Network," in IEEE Access, vol. 7, pp. 167260-167277, 2019, doi:
10.1109/ACCESS.2019.2953542.
[29] Jiang, Weiwei. “Applications of deep learning in stock market prediction: recent
progress.” Expert Syst. Appl. 184 (2021): 115537.
40
[30] Mokhtari, Sohrab & Yen, Kang & Liu, Jin. (2021). Effectiveness of Artificial Intelligence in
Stock Market Prediction based on Machine Learning.
[31] N. Naik and B. R. Mohan, "Novel Stock Crisis Prediction Technique—A Study on Indian
Stock Market," in IEEE Access, vol. 9, pp. 86230-86242, 2021, doi:
10.1109/ACCESS.2021.3088999.
[32] J. Lee, R. Kim, Y. Koh and J. Kang, "Global Stock Market Prediction Based on Stock Chart
Images Using Deep Q-Network," in IEEE Access, vol. 7, pp. 167260-167277, 2019, doi:
10.1109/ACCESS.2019.2953542.
[33] K. Huang, X. Li, F. Liu, X. Yang and W. Yu, "ML-GAT:A Multilevel Graph Attention
Model for Stock Prediction," in IEEE Access, vol. 10, pp. 86408-86422, 2022, doi:
10.1109/ACCESS.2022.3199008.
[34] C. Zhang, N. N. A. Sjarif and R. B. Ibrahim, "Decision Fusion for Stock Market Prediction:
A Systematic Review," in IEEE Access, vol. 10, pp. 81364-81379, 2022, doi:
10.1109/ACCESS.2022.3195942.
[35] S. Saha, J. Gao and R. Gerlach, "Stock Ranking Prediction Using List-Wise Approach and
Node Embedding Technique," in IEEE Access, vol. 9, pp. 88981-88996, 2021, doi:
10.1109/ACCESS.2021.3090834.
[36] Yuan, Xianghui & Yuan, Jin & Jiang, Tianzhao & Ain, Qurat Ul. (2020). Integrated Long-
Term Stock Selection Models Based on Feature Selection and Machine Learning Algorithms for
China Stock Market. IEEE Access. 8. 1-1. 10.1109/ACCESS.2020.2969293.
[37] S. S. Alotaibi, "Ensemble Technique With Optimal Feature Selection for Saudi Stock
Market Prediction: A Novel Hybrid Red Deer-Grey Algorithm," in IEEE Access, vol. 9, pp. 64929-
64944, 2021, doi: 10.1109/ACCESS.2021.3073507.
[38] Wang, Yujie & Liu, Hui & Guo, Qiang & Xie, Shenxiang & Zhang, Xiaofeng. (2019).
Stock Volatility Prediction by Hybrid Neural Network. IEEE Access. 7. 1-1.
10.1109/ACCESS.2019.2949074.
[39] M. Nabipour, P. Nayyeri Z. D. Akşehir and E. Kiliç, "How to Handle Data Imbalance and
Feature Selection Problems in CNN-Based Stock Price Forecasting," in IEEE Access, vol. 10, pp.
31297-31305, 2022, doi: 10.1109/ACCESS.2022.3160797.
[40] L. Zhang, N. Liu and P. Yu, "A Novel Instantaneous Frequency Algorithm and Its
Application in Stock Index Movement Prediction," in IEEE Journal of Selected Topics in Signal
Processing, vol. 6, no. 4, pp. 311-318, Aug. 2012, doi: 10.1109/JSTSP.2012.2199079.
[41] S. Wei, S. Wang, S. Sun and Y. Xu, "Stock Ranking Prediction Based on an Adversarial
Game Neural Network," in IEEE Access, vol. 10, pp. 65028-65036, 2022, doi:
10.1109/ACCESS.2022.3181999.
[42] X. Zhang, S. Qu, J. Huang, B. Fang and P. Yu, "Stock Market Prediction via Multi-Source
Multiple Instance Learning," in IEEE Access, vol. 6, pp. 50720-50728, 2018, doi:
10.1109/ACCESS.2018.2869735.
[43] Z. D. Akşehir and E. Kiliç, "How to Handle Data Imbalance and Feature Selection Problems
in CNN-Based Stock Price Forecasting," in IEEE Access, vol. 10, pp. 31297-31305, 2022, doi:
41
10.1109/ACCESS.2022.3160797.
[44] E. Koo and G. Kim, "A Hybrid Prediction Model Integrating GARCH Models With a
Distribution Manipulation Strategy Based on LSTM Networks for Stock Market Volatility," in
IEEE Access, vol. 10, pp. 34743-34754, 2022, doi: 10.1109/ACCESS.2022.3163723.
[45] S. Kim, S. Ku, W. Chang and J. W. Song, "Predicting the Direction of US Stock Prices
Using Effective Transfer Entropy and Machine Learning Techniques," in IEEE Access, vol. 8, pp.
111660-111682, 2020, doi: 10.1109/ACCESS.2020.3002174.
[46] X. Hou, K. Wang, C. Zhong and Z. Wei, "ST-Trader: A Spatial-Temporal Deep Neural
Network for Modeling Stock Market Movement," in IEEE/CAA Journal of Automatica Sinica, vol.
8, no. 5, pp. 1015-1024, May 2021, doi: 10.1109/JAS.2021.1003976.
[47] J. -S. Chou and T. -K. Nguyen, "Forward Forecast of Stock Price Using Sliding-Window
Metaheuristic-Optimized Machine-Learning Regression," in IEEE Transactions on Industrial
Informatics, vol. 14, no. 7, pp. 3132-3142, July 2018, doi: 10.1109/TII.2018.2794389.
[48] D. Lien Minh, A. Sadeghi-Niaraki, H. D. Huy, K. Min and H. Moon, "Deep Learning
Approach for Short-Term Stock Trends Prediction Based on Two-Stream Gated Recurrent Unit
Network," in IEEE Access, vol. 6, pp. 55392-55404, 2018, doi: 10.1109/ACCESS.2018.2868970.
[49] L. -X. Wang, "Fast Training Algorithms for Deep Convolutional Fuzzy Systems With
Application to Stock Index Prediction," in IEEE Transactions on Fuzzy Systems, vol. 28, no. 7, pp.
1301-1314, July 2020, doi: 10.1109/TFUZZ.2019.2930488.

[50] S. Chen and C. Zhou, "Stock Prediction Based on Genetic Algorithm Feature Selection and
Long Short-Term Memory Neural Network," in IEEE Access, vol. 9, pp. 9066-9072, 2021, doi:
10.1109/ACCESS.2020.3047109.

42