Enhanced Agriculture Forecasting

Using Deep Learning Models

Dr. Muruganandham S1, a) Srividhya S2, b) Varalakshmi K C3, c) Suwathi S4, d)

Selva Bharathi.B5, e) Saranya K6, f)
Authors Affiliations
1
Assistant Professor (Grade-1)
1,2,3,4,5,6
B. Tech Computer Science and Business Systems
1,2,3,4,5,6
Panimalar Engineering College

Authors Emails
a)
Dr.Muruganandham S: [email protected]
b)
Srividhya S: [email protected]
c)
Varalakshmi K C: [email protected]
d)
Suwathi S: [email protected]
e)
Selva Bharathi B: [email protected]
f)
Saranya K: [email protected]

Abstract. Agriculture plays a pivotal role in India, contributing significantly to the nation's economy and day
to day life. In this context, accurate crop yield prediction becomes imperative for informed decision-making
and sustainable resource management. This study focuses on advancing crop yield prediction through deep
learning models, addressing the shortcomings of conventional methods. We introduce a hybrid approach that
integrates Long Short-Term Memory (LSTM) networks, Artificial Neural Networks (ANN) and autoencoders
to enhance prediction accuracy. The existing models often struggle to capture intricate temporal dependencies
and latent features inherent in agricultural datasets. Our proposed model harnesses the sequential memory
capabilities of LSTM and the feature learning process of autoencoders. Through extensive evaluation on
diverse crop datasets, our model demonstrates superior accuracy compared to traditional methods, offering a
robust solution for India's dynamic agricultural landscape. This research contributes to precision agriculture,
empowering farmers with a reliable tool for optimized resource utilization and improved crop yield outcomes

Keywords: Deep Learning, Agriculture Forecasting, Crop Yield Prediction, Long Short-Term Memory
(LSTM), Artificial Neural Networks (ANN), Autoencoders, Temporal Dependencies, Precision Agriculture,
Machine Learning Algorithms, Data-Driven Agriculture

I.INTRODUCTION
Agriculture, India's largest economic sector, is the backbone of the nation's socioeconomic structure,
contributing significantly to the GDP (approximately 17-18%). The escalating demand for production faces
challenges from evolving technologies and over-reliance on non-renewable energy. Changes in rainfall and
temperature patterns, exacerbated by global warming, challenge precise predictions, impacting agricultural
productivity. Crop yield prediction, crucial for optimal outcomes, involves estimating output based on historical
data, encompassing variables such as temperature, humidity, pH, precipitation, and crop type. The integration of
machine learning in agriculture offers potential for revolutionary change by influencing revenue through optimal
crop selection. Recognizing the importance of reliable knowledge, our study employs various deep learning
approaches to estimate crop production, aiming to address multifaceted challenges faced by farmers and the
industry for more informed decision making. In addressing the challenges faced by agriculture, crop yield
prediction becomes a pivotal focus.

Crop yield prediction in agro-meteorology is vital for a nation's economic well-being ,shaping agricultural
landscapes with key crops like jowar, bajra, wheat, rice, and mung. Yield prediction methods, categorized into
statistical and crop simulation models, include notable examples like CERES. Leveraging advanced technologies
such as Machine Learning algorithms and data analytics is transforming the field, surpassing traditional
approaches. Our mission centers on pioneering multiple crop production prediction models, utilizing cutting-
edge deep learning techniques. Through a strategic fusion of AI and climate factors, our system enables farmers
to optimize crop output estimates with precision. This approach aids users in identifying factors influencing crop
productivity, contributing to a nuanced understanding of agriculture. The adaptability and accuracy of "Deep
Learning" align seamlessly with the dynamic nature of agriculture.
Further we depict the results through visual aids like bar charts, pie charts, or graphs, providing farmers with an
accessible way to comprehend and apply the gained insights.

II. LITERATURE SURVEY

[1] The authors describe a systematic method for thoroughly reviewing the use of deep learning in crop yield
prediction. The study employs a well-defined methodology that encompasses a systematic literature review (SLR) to
examine the primary motives, obstacles, employed algorithms, and assessment techniques. This study lays a solid
foundation for improving crop output prediction through Deep Learning by highlighting the need of addressing
obstacles ,improving data availability, and investigating Explainable Artificial Intelligence (XAI)for model
transparency.

[2] The authors of the study provide a thorough literature review on the use of hybrid deep learning- based models
for crop yield prediction. The research discusses the challenges involved in yield prediction and assesses the
effectiveness of a number of algorithms, such as CNN-XG Boost, CNN-Recurrent Neural Networks (RNN), CNN-
XGBoost, CNN-Long Short Term Memory(LSTE),and Convolutional Neural Networks (CNN). The study's best
performing model, the brid CNN-DNN model, is shown on a soybean dataset. issues with predicting crop yield and the
significance of precise forecasts for food supplies.

[3] The authors here employ deep learning and machine learning models, such as Random Forest Support Vector
Machine (SVM), Lasso Regression, Gradient Descent, and Long Short-Term Memory (LSTM), to estimate agricultural
productivity. In the analysis of the effects of various parameters on model accuracy, RandomForest outperforms other
models in terms of crop yield prediction. The study highlights the value of complete datasets with precise forecasts and
makes suggestions for future research extensions

[4] The authors employ deep learning techniques like CNN, DNN, and LSTM for crop yield prediction, leveraging
factors such as climate, soil, and crop images. CNN achieves 82% accuracy in classifying food crops. Two-layered
CNN for paddy fields attains 86.5%.DNN exhibits 64% accuracy, and Two-layered DNN achieves 67%. LSTM,
notably, reaches an accuracy of 86.3%, while Two-layered LSTM excels with an accuracy of89%. Hybrid networks,
combining neural networks with machine learning, attain accuracies ranging from 85%

[5] The authors of use several deep learning methods in their research to estimate yield. There is discussion of
machine learning methods such as neural network, decision trees. And support vector machines, with an emphasis on
deep learning models like Recurrent Neural Networks(RNN) and Convolutional Neural Networks (CNN). The
development of Gated Recurrent Units (GRU)and Long Short-Term Memory (LSTM) structures results from the
resolution of the RNN's restrictions. Additionally, the Bidirectional LSTM (BLSTM) model is generated, which
improves outcomes by taking into account contextual data from both ends of a sequence.

[6] The authors are aimed to assess the accuracy and interpretability of deep learning models for crop yield
forecasting. Three models were constructed: linear trend models (baseline), Gradient-Boosted Decision Trees (GBDT)
with expert-designed features, and (LSTM,1DCNN) automatically extracting features. Feature attribution methods
were employed for interpretability, validated by human experts. Evaluation metrics included normalized root mean
squared error (NRMSE), and model performance was compared using statistical tests and spatial variability analysis in
France.

[7] The authors outlines the data preprocessing steps, incorporating genotype data coding and weather prediction
through neural networks. The authors employ deep neural networks for yield prediction, comparing their approach
with Lasso, shallow neural networks, and regression trees. The findings emphasize the effectiveness of deep neural
networks in crop yield prediction.

[8] The authors create an integrated Internet of Things(IoT)-based agricultural system that includes AI- based
rainfall prediction, real-time data monitoring, irrigation control, and fruit health detection. For better performance,
they incorporate CNN, LSTM, and attention layers using Node MCU, a variety of sensors, and an Android app. The
system's usefulness is demonstrated through experimental validation.

[9] The authors Johnathon S., Gangopadhyay T.,WuL. Ganapathy ., Sarkar S., and Singh A. K. outperformed
baselines and the USDA model in predicting seed production. Their research demonstrated enhanced efficacy and
comprehensibility in identifying temporal correlations for
agricultural forecasts.

[10] In order to predict the Kharif crop yield, the authors UsharaniBhimavarapu, Gopi Battineni, and Nalini
Chintalapudi used Indian data with CNN, RNN, GRU, and LSTM models with an Improved Optimization
Function(IOF). Among the performance measurements wereR2,RMSE,MAE, and MAPE. The goal of the study was
to reduce loss by adjusting the IOFLSTM model's optimization parameters.

III. EXISTING SYSTEM

The current crop yield prediction technique relies on instance-level approaches utilizing deep neura networks to
transform independent instances into embedding space. However, this method lacks automation in recognizing crop
proposals and behavior analysis, making it suitable only for specific instance classifiers. The existing crop yield
prediction system incorporates various deep learning models such as Convolutional Neural Networks (CNN), Deep
Neural Networks (DNN), Long Short-Term Memory (LSTM), 1D Convolutional Neural Network (1DCNN),
Bidirectional LSTM (BLSTM), and Gradient- Boosted Decision Trees (GBDT). These models aim to enhance
accuracy and interpretability in yield forecasting, yet challenges persist in mitigating the impact of climatic
uncertainties on their performance. There is a need for more advanced and automated methodologies to enhance the
overall effectiveness of crop yield prediction using deep learning models.

IV. PROPOSED WORK

The developed model uses a two-phase training and testing strategy to forecast the top ten crops in a particular
Tamil Nadu district. Training involves gathering data and applying a variety of deep learning techniques—such as
autoencoders, ANNs, and long short-term memory(LSTM)—to increase accuracy. This combination ensures
effective feature representation, analyzes sequential data, and captures non-sequential patterns in the best possible
way. A comprehensive comprehension of crop production dynamics is ensured by utilizing autoencoders for feature
representation and LSTM for sequential analysis. The input, ANN, LSTM, and autoencoder layers in the suggested
model's architecture work together to optimize their capabilities for precise temperature, rainfall, and crop
production forecasts—especially in Tamil Nadu's ever-changing agricultural environment

V. METHODOLOGY

5.1 Long Short-Term Memory

Crop yield prediction with Long Short-Term Memory (LSTM) involves harnessing the power of recurrent neural
networks to forecast agricultural output. LSTM models excel in capturing temporal dependencies and patterns within
multivariate time- series data, such as weather conditions, soil attributes, and historical yields. By leveraging these
factors, LSTM networks can predict future crop yields with a high degree of accuracy. This technology empowers
farmers and stakeholders with valuable insights into potential harvests, facilitating proactive decision-making,
resource optimization, and overall enhancement of agricultural productivity and sustainability

LSTMs are excel at handling temporal sequences. In crop yield prediction, historical data related to crop
yields, weather conditions, and other relevant factors are organized into sequences. LSTMs can effectively learn
patterns and dependencies within these sequences. The non- linear nature of LSTMs enables them to capture
complex relationships between input variables. In crop yield prediction, where the interactions between
environmental factors are often non-linear, LSTMs can provide more accurate predictions compared to linear
models. LSTMs can automatically learn complicated patterns in sequential data, retain long-term memory, and
handle temporal relationships effectively, they are a great tool for predicting agricultural yield. Their versatility in
handling disparate datasets and incomplete data makes.

5.2 Artificial Neural Network

In the context of predicting crop yields, the implementation of Artificial Neural Network
(ANN)incorporated regression methodologies for yield estimation. The model integrated diverse input parameters,
encompassing images, soil properties, and environmental factors. Periodic scrutiny of crop images, focusing on
growth stages, height, diseases, and overall development, proved instrumental in yield prognostication.

Environmental factors like temperature, rainfall, and irrigation were factored in to gauge crop sustainability.
The trained ANN model, utilizing data from wheat, barley, sugar beet, potato, and sunflower crops, achieved an
average estimation accuracy ranging between 60% and 70%. To refine yield predictions, multiple linear regression
(MLR)was employed, yieldinga67%accuracyby analyzing temperature, humidity, and moisture at regular intervals.
Addressing specific soil characteristics for each crop, a more efficient strategy incorporated K-means Nearest
Neighbor (KNN) clustering, grouping
 5.3 Autoencoders
In crop yield prediction, autoencoders are used to identify significant characteristics and decrease the dimensionality
of input data, improving the prediction model's efficiency. Because they process and extract critical information
from high-dimensional agricultural datasets in an effective manner, autoencoders are vital to the prediction of crop
output. Autoencoders are used in crop forecasting to reduce dimensionality and eliminate noise, which improves the
model's capacity to identify significant patterns in input variables like soil and weather information.

Autoencoders are essential for temperature and rainfall forecasting in addition to crop prediction. These models aid
in precise weather forecasting by analyzing and identifying significant characteristics from the input data. Because
autoencoders can recognize important patterns, they are useful tools for many different applications, one of which is
weather forecasting.
VI. TIME COMPLEXITY

The time complexity of LSTM, Autoencoder, and ANN for crop prediction varies with factors like model
architecture, dataset size, and training iterations. LSTMs, with sequential processing, may have higher complexity,
while Autoencoders and ANNs depend on layer sizes. Dataset characteristics and training parameters further
influence overall complexity.
The time complexity for prediction tasks in LSTM ,Autoencoder and ANN can be denoted as
(O(f (N, M, P))
Where,
N denotes the data set size,
M denotes model
parameters,
P denotes the number of training iterations.
These denotation captures the dependence on key factors influencing the computational time for the given
prediction task.

VII. SYSTEM DESIGN

Figure1. Diagrammatic representation of System

 VIII. IMLEMENTATION
The previous section detailed the agricultural harvest yielding system's architecture and technology stack. This section
covers the implementation and operation of the system. The system's output and results are shown in the following
figures.

FIGURE 1

As per figure 1, it shows the comparison between predicted and actual yield production for different crops

FIGURE 2

The wheat crop's actual yield during a ten-year period is compared to its projected yield in this point graph, which is
displayed in Figure 2. It is evident that the projected production is more accurate than the actual production and
occasionally falls inside a safety margin.
 FIGURE 3
The figure 3 intricately depicts the average rainfall throughout many districts, providing a thorough visual depiction that
distinguishes the differences in precipitation amounts within each.

FIGURE 4

The above figure 4 clearly depicts the average temperature throughout numerous districts, offering a thorough visual
description of the temperature dispersion throughout the area. This graphical representation provides useful insights into
localized climatic patterns, allowing for a better understanding of how temperature varies throughout different
geographical areas within the districts.
 FIGURE 5

The figure 5 depicts crop yields and planting areas across districts in a given year. It covers a wide range of crops,
including rice, maize, bajra, ragi, gram, toor, mung, and jowar. This detailed analysis provides information on regional
agriculture productivity and cultivation practices. It is an invaluable resource for politicians, researchers, and farmers who
want to make informed decisions. The data is useful for planning crop strategy, allocating resources, and implementing
agricultural development projects.

VI. CONCLUSION AND FUTURE ENHANCEMENTS

This research explores the revolutionary effects of deep learning in agriculture, concentrating on the temperature, rainfall,
and crop yield predictions for ten different crops in Tamil Nadu. Using an efficient combination of ANN, autoencoders,
and LSTM, our model demonstrated impressive accuracy, with scores ranging from 85% to 95%. The incorporation of
these advanced algorithms was crucial in interpreting complex patterns seen in the agricultural data.

In the future, we're dedicated to improving forecast accuracy by incorporating state-of-the-art technology and expanding
the variety of datasets we use. Subsequent endeavors entail delving into hardware integration, with the aim of enhancing
the system's efficiency and furnishing farmers with up-to-date information. By providing farmers with thorough
information to enable them to make educated decisions, this all-encompassing strategy seeks to promote resilient and
sustainable farming methods.

The study foreshadows a paradigm change in precision agriculture by emphasizing the significant advantages of
technologically driven solutions. Our commitment to improving prediction accuracy and developing novel approaches to
improve this system's usability and accessibility for farmers and society at large does not waver as we look to the future.
This study is a critical first step toward a future in Tamil Nadu agriculture and beyond that is more data-driven and
technologically integrated.

VI. REFERENCES

[1] Nikhil, Dr.Kanwal, “Exploiting Data Mining Technique for Rainfall Prediction,” International
Journal of Computer Science and Information Technologies (IJCSIT),vol. 5 ,no 3,pp.3982-3984,2019

[2] Kumar, Y. J. N., Spandana, V., Vaishnavi, V. S., Neha, K., & Devi, V. G. R. R. . Supervised
Machine learning Approach for Crop Yield Prediction in the Agriculture Sector. In 2020 5th International
Conference on Communication and Electronics Systems (ICCES) {pp. 736-741}, IEEE,2020

[3] Patil, A., Kokate, S., Patil, P., Panpatil, V., & Sapkal, R. Crop Prediction using Machine Learning
Algorithms. International Journal of Advancements in Engineering & Technology, 1{1}, 1-8-2020

[4] Tanhim Islam, Tanjir Alam Chisty, BRAC University, A Deep Neural Network Approach for Crop
Selection and Yield Prediction in Bangladesh(2018)
 [5] Talha Siddique, Dipro Barua, Zannatul Ferdous, Amitabha Chakrabarty Department of Computer
Science and Engineering BRAC University, Automated Farming Prediction

[6] Alexandros Oikonomidis, Cagatay Catal and Ayalew Kassahun, Deep learning for crop New
Zealand Journal Of Crop And Horticultural Science 2023, VOL. 51, NO. 1, 1–26

[7] Kavita Jhajhariaa . Crop Yield Prediction using Machine Learning and Deep Learning Techniques
in International Conference on Machine Learning and Data Engineering, Manipal University Jaipur(2023)

[8] M K Dharani , R Thamilselvan , P Natesan ,PCD Kalaivaani , S Santhoshkumar,

Review on Crop Prediction Using Deep LearningTechniques, Journal of Physics: Conference Series 1767 (2021)

[9] Khaki L and Wang L, 2019, Crop Yield Prediction Using Deep Neural Networks,Front. Plant Sci.,10

[10] Schwalbert R A, Amado T, Corassa G, Pott L P, Prasad P V V and Ciampitti I A, 2020, Satellitebased
soybean yield forecast: Integrating machine learning and weather data for improving crop yield prediction in
southern Brazil, Agric. For.Meteorol.,284.

[11] Alexandros Oikonomidisa, Cagatay Catalb, and AyalewKassahun, Hybrid Deep Learning-based Models for
Crop Yield Prediction Applied Artificial Intelligence 2022, VOL36,NO.1,e2031823

[12] Khadijeh Alibabaei ,Pedro D. Gaspar and Tânia M.., Crop Yield Estimation Using Deep Learning
Based on Climate Big Data and Irrigation Scheduling, . Energies2021,14,3004,Published:22May,2021.

[13] Dilli Paudel , Allard de Wit , Hendrik Boogaard , Diego Marcos , Sjoukje Osinga , IoannisN.Athanasiadis,
Interpretability of deep learning models for crop yield forecasting,
Computers and Electronics in Agriculture
206 (2023) 107663

[14] van Klompenburg, T., Kassahun, A., Catal, C., 2020. Crop yield prediction using machine learning: A
systematic literature review. Comput. Electron. Agric. 177,105709

[15] Khaki S and Wang L (2019) Crop Yield Prediction Using Deep Neural Networks. Front. Plant Sci. 10:621.
doi: 10.3389/fpls.2019.00621

Cmaak Z.,B.,,Nagulla B..,Ponduru B.,Sowmya V, Rainfall Prediction using Machine Learning & Deep Learning
[16]
Techniques, 2020 International Conference on Electronics and Sustainable Communication Systems (ICESC)

[17] Akash M., Saikat B.., Effective Crop Prediction Using Deep Learning, 2021 International Conference on Smart
Generation Computing, Communication and Networking (SMART GENCON)

[18] Dhivya Elavarasan , P.M Durairaj V.,, Crop Yield Prediction Using Deep Reinforcement Learning Model for
Sustainable Agrarian Applications IEEE Access ( Volume: 8)
2020
[19] Pragneshkumar P.,, Sanjay C.., Hasit P.., Analyze the Impact of Weather Parameters for Crop Yield Prediction Using
Deep Learning, International Conference on Big Data AnalyticsBDA 2022: Big Data Analytics pp 249–259.

[20] Kandi V. N. R. and Evuri B. R., Crop Yield Prediction Based on Weather and Soil Parameters Using Regression
Tree Model, International Conference on Communication, Devices and Computing,ICCDC 2023: Proceedings of the
4th International Conference on Communication, Devices and Computing pp 1–10

[21] P. Sharma, P. Dadheech, N. Aneja and S. Aneja, "Predicting Agriculture Yields Based on Machine Learning Using
Regression and Deep Learning," in IEEE Access, vol. 11, pp.
111255-111264, 2023, doi: 10.1109/ACCESS.2023.3321861.

[22] Amit K. S., Nima S., Saeed Khaki, Gina L., Wenzhi Zeng, Frank Ewert,Thomas G., & Jaber R., Winter wheat
yield prediction using convolutional neural networks from environmental and phenological data,
Article number: 3215 (2022)