Ctr ere Model Question Paper - 2 ‘Time: 2% Hours Instructions : Answer All Sections: 1, Whatis supery vised Macl 2. Why Python is used for Machine Learning? 3. What is Data Preparation ? 4. What is Regression? 5. What is Diserete Output Variable? Give an example. five an example. ean example ee ee Rrerenernen no een 7. What is Unsupervised Machine Learning? Explain the Key Components of Unsupervised Machine Learning. . What is SciPy? Why itis needa for ML? Explain its features. . How to Handle Missing Values and Ouliers? Explain with an example. Explain the Process of Getting the Data. plain the Differences between Regression and Classification. Explain the Limitations of K-Means Clustering. Bee eos et 13. Explain the Main Challenges of Machine Learning 14, How Semi-Supervised Machine Learning Works? Explain with an example, 15. a) How to Create a Test Set? b) Why Data Reduction is Important in ML2 16. a) What is Logistic Regression? Explain how it works? b) Writea Python Code for Spam Email Detection using the Naive Bayes classification algorithm, 17. a) Write Decision ‘Tree Algorithm and explain how it works? b) Explain how a cluster formed in DBSCAN clustering algorithm? 18. a) Explain the types of Clustering Methods or Techniques. b) Write a Python Code to Use Clustering in Semi-Supervised Learning e000‘Supervised machine learning algorithms are Intendee' Sv led to the machine learning system for trainin, supervised learning, sé belled data are provid ie ea i data, The system uses labelled dy cd the system then predicts the output based on the training : to build a model that understands the datasets and learns about each one. After the training an ins processing are done, we test the model with sample data to see ifit can accurately predict the outpu Qf tect gt $<] ne ‘a Now the first of + Supervise. earning isa type of machine learaing where the algorithm leayns to map input data to eee the correct output by being trained on labeled examples. In supervised learning, the training data consists of input output pairs, where the input data is accompanied by the corresponding correct output or label. The goal of supervised learning is for the algorithm to learn a mapping function from the input to the output sc that it can make accurate precictions on new, unseen data. Cheouut eet 4 in supervised learning, the training data is labeled, meaning that each inp! * Testin data point is associated with the correct output or target label. For example, in a spam emo atl detection system, each email is labeled as either spam or not spam. Set (eg me ; . : ra ‘Training Phase: During the training phase, the algorithm uses the labeled data to learn th | i “htlnship between the input features and the corresponding output labels, The algorith| |” Pd “ists its internal parameters based on the training data to minimize the o ror between it criteria predictions and the true labels. bleak HWitha, 4. Testing Phase: Once the mode! : . a : One! ode! is trained, itis tes : Ev, 's tested on a separate test dataset that contal ‘lua ‘new data that it has not seen hefare8 SENSSHING Learning KCB essence, the model is not able to learn enough from the training exampl B examples to my [ data. b accurate predictions. To overcome this issue: © Maximize the training time © Enhance the complexity of the model \ © Add more features to the data | © Reduce regular parameters © Increasing the training time of model Example: Let's consider a scenario where a linear model is used to predict housing pric based solely on the number of bedrooms in a house. If the model is underfitting, it ms oversimplify the relationship between the number of bedrooms and the price, assuming tha | all houses with the same number of bedrooms have identical values. This simplistic approach | overlooks other crucial factors influencing housing prices, such as location, square feet, and | amenities. TE Why Pytho. Python is use* in numerous data science and machine learning applic user-friendly nature. It seamlessly combines the robust capabilities of general-purpose program NS tanguages with the simplicity of domain-specific scripting languages like MATLAB or Witharicy Ps ecosystem of libraries catering to data loading, visualization, statisties, natural language processing entists with a diverse toolkit encompassing. both al image processing, and more, Python equips data sci general and specialized functionaiities. ‘The Nexibility of Python extends to its interactive nature, allowing users to engage directly with the code through interfaces like terminals or popular tools such as Jupyter Notebook. This interactive capability is particularly advantageous in the iterative nature of machine learning and data ansiysis, where insights are derived from the data itself. Python has emerged as the preferred language for machine learning due to its exceptional qualities. Python is renowned for its simplicity, readability, and extensive library ecosystem tailored for data science and machine learning tasks. Libraries like NumPy, Pandas, Scikit-learn, TensorFlow, and PyTorch provide powerful tools for data manipulation, visualization, and model building. gem Python is Preferred Choice for Machine Learning Appli | Python is the preferred language for machine learning due to several key reasons, rw ations due to its versatilityand Se ns? | . a poeaee of Libraries: Python includes a vast array of libraries and frameworks specitically reated for machine learning and data science tasks. Libraries like NumPy, Pandas, Scikit-learn, | ‘TensorFlow, and PyTorch provide powerful tools for data manipulation, visualization, and building , I tool: 8 | elearning models + Ease of Lea ; | irning and Use: Python is popular for its simplicity and readability. Its clean syntax and extensive documentation enabl le devel io L Poe lopers to write code efficiently, accelerating the development+ Example: Following training an id validatio, ew to the model with undisclos, a @ fresh dataset ed selling as Prices, is reserved Testing data provides an unbiased enn? (2 5 Valuation "id scenarios, indicating it capacity to ge model's performance in real-w accurately to new instances. 1 data helps tune the ‘on unseen instances. Each workflow, ensuring that the model is ce. and capable of making accurate predictions on ‘ype of data serves a specific purpose in the Machine Learning, trained effectively, optimized for performan i ata Preparation in Machine Learning Data preparation is a critical step in the machine learning pipeline that involves processin; transforming raw data into a format suitable for building and training machine learning models. This process ensures that the data is clean, relevant, and structured in a way that optimizes the performance of machine learning algorithms. Each machine learning project requires a specific data format. To do so, datasets need to be prepared well before applying it to the projects. Sometimes, data in data sets have missing or incomplete information, which leads to less accurate or incorrect predictions. Further, sometimes data sets are clean but not adequately shaped, such as aggregated or pivoted, and some have less business context Hence, after collecting data from various data sources, data preparation needs to transform raw data What is Data Preparation ? + Data preparation is defined as a gathering, combining, cleaning, and transforming raw data to make accurate predictions in Machine learning projects. Itis the later stage of the machine learning lifecycle, which comes after data collection.Supervised Learning Regression is a process of finding the correlations between dependent and independent varia It helps in predicting the continuous variables such as prediction of Market Trends, prediction House prices, etc. The task of the Regression algorithm is to find the mapping function to Tap ¢ input variable(x) to the continuous output variable(y). Regression algorithms are used if there is a relationship between the input variable and the outpy variable. [tis used for the prediction of continuous variables. The goal of regression tasks is to predi a continuous numbe or a real number. If there is continuity between possible outcomes, then th problem is a regression problem. r What is Regression im LQ © + Regression in machine learning is a type of supervised learning problem where the goal is to predict continuous numerical values based on input features. Unlike classification, which predicts | discrete class labels, regression models estimate a continuous output variable. The objective of | regression is to establish a relationship between the input variables and the continuous target | variable and allowing the model to make predictions on new or unseen data points. | Example : Predicting house prices, forecasting stock prices, estimating the temperature based on weather variables. \ 3 [Examples [Regression Tasks | 4. Real Estate Valuation: | + Task: Predict the market value of properties. «Features Features like area, numivcr of bedrooms, age of the property, amenities, etc. Estimated price of the property. + Purpose: Helps buyers and sellers geta fair idea of property prices and assists investors and rea | estate companies in making informed decisiorss. 2. Stock Market Prediction: + Task: Predict the stock price or return value « Features Historical stock prices, trading volume, market indices, economic indicators. + Target Variable: Estimated Future stock price or return \ + Purpose: Using regression analysis, historical stock data, and relevant market indicator + g informe model can predict the future price or return of a stock. It helps investors in making ink trading decisions. 3. Credit Risk Assessment: + Task: Predict the probability of default or credit risk + Features: Credit score, income, debt-to-income r io, loan amount. + Target Variable: Probability of default or credit risk 7 1 borrow? . , a i alyzini + Purpose: Regression models can be employed to assess credit risk by analyzing 101s nee or dafault or the level of erttasks icting vature re the ‘input quires abels), lasses arious isions 2 L tet lable is, die Malis spam (class 1) or not two classes, ible is discrete, taki ; two distinct values (Oar I represenroe ey Classifying Images of Fruits Classiving images of fruits inn He (class 0) OF Lor 2) ee TMB (4a 2) The eupat sne e vith multiple distinct vales (0 or2) representing the three class. ; (apple, banana, orange) + Sentiment Analysis: Analy: neutral). The sentiment lab text. aie {ext dita to determine the sentiment ofa review (positive, negative, ls (positive negative, neural) are discrete categories assigned tthe input Medical Diagnosis: Predicting the presence ofa disease based on patient symptoms and test results The diagaosis categories (eg, disease presen, no disease) are discrete labels assigned to the patient data, Custemer Retentios Prediction). Predicting whether a customer will renew a subscription or not (churn Discrete Output Variables: The term "discrete" in the context of machine learning refers to 3 type of variable that has specific and separate values, as opposed to continuous variables, which can take any value within a range, Discrete variables are countable and have distinct categories or values, which cannot be subdivided meaningfully. Examples of Discrete Outputs : + Gender Classification: Male, Female + Loan Approval: Approved, Rejected + Movie Genre Classification: Action, Romance, Thriller, Comedy, Drama Supervised Learning: Classification is a supervised learning approach, meaning it relies on labeled training data to learn the relationship between input features and the target classes.KEE types oF Unsupervised bear Unsupervised learning includes a divers ny, 1 of techniques airned and insights trom datasets consis fof input data without unsupervised Ie, upervised Teaming methods are listed below: 1. Clustering: Clustering is the process of grouping a set of objects « that objects in the same group (called a cluster) are more sim in other groups. It's often used in e 5 to find natural ploratory data analy! patterns and outliers, and simplify complex data sets. Common Clustering Algorithms: + K-Means Clustering: Divides a set of samples into disjoint © the mean of the samples in the cluster. + Hierarchical Clustering: Builds a multilevel hierarchy of clusters © tree + DBSCAN (Density-Based Spatial Clustering of Applications with Noise) clusters as areas of high density, allowing it to find arbitrarily shaped < more robust to outliers than K-means. « Gaussian Mixture Models (GMM): Models clusters as a mixture of distributions. Points are probabilistically assigned to clusters, which clustering. 2. Association : Association is arule-based machine le. ing method used to discove relations between variables in large databases I's often used in marker baske jee, zeveal combinations of products frequently bought together Common Association Algorithms: | Apriori Algorithm: Identifies frequentindividual tems in the database and ext ta larger and larger item sets as long 2s those item sets appear sufficiently frequ the database «_ pP-Growth: Used for finding frequentitem sets in a dataset for 3, Dimensionality Reduction: Dimensionality reduction methods aim to reduce the nu features in a dataset while preserving as much relevant information as possible. Common Dimensionality Reductio! Techniques: "principal Component Analysis (PCA): A statistical procedure that uses an orthogona transformation to convert a set of observations of possibly correlated variables into 3 of values of linearly uncorrelated variables. | {Distributed Stochastic Neighbor Embedding ((-SNE): A tool to vist dimensional data by reducing it to two or three dimensions while maintainin ances between points. . 8 a coders: Neural networks use earning effie Autoen : - ee rks eu for learning efficient codings of unlabel ‘The aim is typically to reduce dimensionality by learning a network \ toa lower-dimensional space and then decodes it back to the acy ‘ae : he 4, Anomaly Detection: Anomaly detection is the ide . observations which are significantly differes ciation rule ed d. ed data encodes the dat original space tification of rare items, ntal i wens g nd deviate from the majority oft th, eee_— ao Machine Learning wd The prim Dnsupervised Learning Ee word unsupevsel implies “not Being done or acting under supervision Unsupervised Te Fog i 2 feating method in which a machine learns without any supervision. The traning s to the machine with the set of data that has not been labelled, classified, or categorized. and thm needs to act on that data without any supervision. ary goal of Unsupervised learning is often to discover hidden patterns, similarities, or ers within the data, which can then be used for various purposes, such as data exploration ‘eualization, dimensionality reduction, and more, Pee Se ee nasa «Unsupervised learning isa type of machine earning where the algorithm learns to identify patterns and relationships in data without being explicitly trained on labeled examples. Unlike supervised Jearning, unsupervised learning algorithms work on unlabeled data, where the algorithm tries to find hidden structures or patterns within the data, The goal of unsupervised learning isto explore the data and extract meaningful insights without the need for predefined labels. pervised learning, the algorithm works with unlabeied data, meaning 1. Unlzbeled Data: in unsuj 's task is to that the input data points do not have corresponding output labels. The algorithm’ discover the underlying structure or patterns in the data on its owr. 2, Clustering: Clustering is a common technique in unsupervised learning where the algorithm groups similar data points together based on their features or characteristics. Clustering algorithms aim to partition the data into clusters such that data points within the same cluster are more similar to each other than to those in other clusters. 3, Dimensionality Reduction: Dimensionality reduction techniques are used in unsupervised learning to reduce the number of features in the data while preserving important information. This helps in visualizing high-dimensional data and removing noise or redundant information. 4, Anomaly Detection: Unsupervised learningalgorithms can also be used for anomaly detection, where the algorithm identifies data points that deviate significantly from the norm or expected behavior. Anomalies are data points that are rare or unusual compared to the majority of the data 5. Association Rule Learning: Association rule learning is another technique in unsupervised learning that discovers interesting relationships or associations between variables in large datasets, It is commonly used in market basket analysis to identify patterns in consumer behavior,_——— eMeStandard deviation of the array, re Presented by mp.std(a), measures the amount of variation or ! dispersion of the set of values, | + Thevariaitce ofthe wvecalculated by npvar(a)is another measure of dispe sion, which is essentially the square of the standard deviation + Them the + The dian found using np.median(a) isthe middle value in the sor higher half from the lower half of the data set Percentile is computed with np percentile data falls. In this case, Hot of numbers that separates (2, 50), which represents the value below which a given the SOth percentile is calculated, which is also known as the | percentage of the median, SciPy is an open-source Python library that is used for scientific and technical computing It bus or of NumP: and provides a wide ange ffnctons for numerical ineraion opinion. si ‘ Neotel iPy is rful tool for scientific computingly, ics, and much more. SciPy isa powerful too ssing, linear algebra, statistics, an cent comp and is widely used in various fields, including machine earning, physies, engineering, and biology and is widely een) Machine Learning 1. Integration and Opt interpolation, and optim ization: SciPy Se ncludes functions lumerical integr gration, zation. These capabilities are essential for problems in machine learning, such as par machines (SVM) or neural networks, 2, Signal Processing: SciPy offers tools for signal processing tasks like filtering, spectral ana y and waveform generation. These functions are valuable for processing and analyzing signatay machine learning applications, such as speech recognition or image pr : solving optimizatio: ‘ameter tuning in algorithms like support vert 4 ‘cessing. Linear Algebra: SciPy provides a comprehensive set of functions for linear algebra operation including matrix decomposition, eigenvalue problems, and solving linear systems of equations These operations are crucial for many machine learning algorithms that involve matrix computations. 4. Statistics: SciPy includes statistical functions for probability distributions, hypothesis testing, and descriptive statistics. These functions are useful for data analysis, model evaluation, and understanding the significance of results in machine learning experiments. 5. Sparse Matrices: SciPy supports sparse matrix representations and provides efficient algorithms for working with large, sparse datasets. Sparse matrices are commonly used in machine learning for tasks like collaborative filtering, text mining, and graph anaiysis. 6. Image Processing: SciPy includes modules for image processing tasks such as filtering, edge detection, and morphology. These functions are beneficial for preprocessing image data in machine learning applications liké computer vision and object r. cognition. 7. Interepcrability with NumPy: SciPy seamlessly integrates with NumPy, making it easy to combine the array manipulation capabilities uf NumPy with the advanced scientific computing functions of SciPy. This integration enhances the efficiency and productivity of machine learning workflows. 1. Scientific Analysis: For tasks that require precise calculations and data analysis, such as in physics and chemistry, SciPy provides robust algorithms that are dependable and efficient es use SciPy for simulating real-world 2. Engineering Applications: Many engineering disciplin processes, optimizing systems, and analyzing data. pn sychology utilize 3. Academic Research: Researchers in fields like economics, sociology, and psycholog SciPy’s statistical tools to analyze experimental data, supports tasks in multi-dimensional it 40 i “s sub-package ndimage Image Processing: SciPy's sub-packag ie ys and computer visio. Processing, widely used in fields such as medical imageDate Preperation * Normalization: Cleaning je dataset for analysis ita helps in standardizing and preparing + Featu re Engineering: Data cleaning is crucial for effective feature extraction and selectior * Mod may Performance: Clean data enhances model performance, reduces errors, and ‘Overall efficiency ML tasks. Re aes Missing values in data refer to the absence of information or data points for certain observations attributes in a dataset, Handling missing values is crucial in data preprocessing to ensure the gus and reliability of the machine learning model. Exampl * The Titanic Passengers dataset has missing values in the Age and Cabin columns Passenger information has been extracted from various historical sources. In this case the m values couldn't be found in the sources [Passengerd] Survived] Pelass | Gender | Age] SibSp| Parch | Ticket Fare [Cabin [Embaried } Z 9 3 | Mal 22{ a | o [a/s2iim 725 2 joa 1 Female, 38 | 1 o | PC 17599 | (71.2833 | CBS a 3__| Female| 26 | 0 | 0 |STON/O2311282|7025 { > 4 [oi tt Female | 35 | 1 0 | 113803 [53.1 C123 i 5 | 0 | 3 | Male | 3s] 0 | 0 [373450 (805) ‘ Lets yy 3 Mae { ) 0) 0 1330877 = Missing values Commun Methods to Handle Missing Values + Deletion: Involves removing entire rows with missing values. While simple, it can lead to | of valuable data. Mean/Median/Mode Imputation: Replace missing values with the mean, median, oF moc! of the vesnective feature. This method is simple but may distort the original distribution + Forward Fill/Backward Fill: Fill missing values with the most recent non-missing value (forward fill) or the next non-missing value (backward fill) along the column. + K-Nearest Neighbors (KNN) Imputation: Predict missing values based on the values of t! nearest neighbors in the feature space. Prediction Models: Use machine learning algorithms to predict missing values based 0 other features in the dataset. This approach can be effective but requires more computation? resources | By Example | Handling Missing Values | Let's consider an example dataset with missing values in the "Age" and “Income” columns: 5 iD [Gender Tncome [Region ‘| Male 50000 [tas 2 [femal 0000 [West | | falar re 1 l a xo° 18 4 DataFrame using the Code snippet uses Simpletmputer trom sik learn | rea Ha-learn tohandte missing vatues in, pandas Datatrame * By setting the imputer strategy to “mean,” missing values it the Dat of each column, Frame are replaced withthe mean * The cusie demonstrates practica "PProach to data cleaning, ensuring the dataset i prepared for) analysis or machine learning tasks, * The fit-transform() metied of the dng spied to fill missing values, resulting ina cleaned DataFrame ready for further Processing | * The code showcases the seamless integration of pandas and scikit-learn for d emphasizing the im tice of handing missing values in ma Outliers are data points that. Significantly differ from other observationsin a datasetThese data points Can skew statistical analyses and machine learning models, leading to inaccurate results. Outliors can occur due to various reasons such as measurement errors, daca entry mistakes, or genuine extreme values in the data, Example : In a dataset containing information about individuals, encounter outliers, such as ages above 100 years, While some in years old, extreme ages can impact statistical analyses and mach appropriately. Identification of Outliers: * Visual methods like box lots, scatter plots, and histograms can help identify outliers. statistical methods such asz-scores, QR (Interquartile Range), and Tukey's method can be used todetect outliers, Such as their age, itis common to dividuals may indeed be over 100 ine learning models if not handledeX ochre tasks ; Common Methods to Handle Outliers : hat are , ee ints tha + Removing Outliers: This method involves identifying and removing data pein aan considered outliers from the dataset. Outliers are detected using statistical me easel z-scores, IQR, or domain knowledge. Once identified, outliers are removed from the dataset ‘o prevent them from affecting the analysis or model training. Removing outliers can le a loss of information, especially if the outliers are valid data points. Careful consideration needed to ensure that the removal of outliers does not bias the analysis. + Transforming Data: Data transfor mation techniques are applied to adjust the distribution of thedataandreduce the impactof outliers. Common transformationsinclude log transformation, Square root transformation, or Box-Cox transformation, These transformations help make the data more normally distributed and lessen the influence ofextreme values, Data transformation can help improve the performance of models that assume normality in the data distribution, * Capping or Winsorizing: Capping involves setting a threshold beyond which values are capped, whjle winsorizing replaces extreme values with a specified percentileCapping and winsorizing are useful when outliers are valid data points but need to be controlled in the analysis. * Binning: Grouping outliers into a separate category or bin can be a suitable approach depending on the nature of the data. * Advanced Models: Utilizing robust models that are less sensitive to ouidiers, such as Random Forest or Support Vector Machines, can be beneficial. of Example “| Hendling Outliers = | the dataset contains salaries ranging from 30,000 to 3,00,000, with a few entrice exceeding 5,00, = | (considered outliers): * Apply capping by setting a threshold atthe 95th percentile (eg, 4,00,000), + Any salary above 400,060 iscapped at 400,000 to prevent extreme values from ske ing the analysi HERE rain ee ——S port pandas as pd ———] import numpy as np # Example DataFrame with potential outliers | data = {'values': (18, 20, 30, 150, 25, 35, 200, 40]) | df = pd.DataFrame(data) | # Calculate the 2-score GF['Z score] = (dF["Values'] - dF{*Values'].mean()) / GFL Values"). std(ddoreoy | # Define a threshold for outliers (commonly set to 1 or 2 or 3) threshold = 1 se eeee a ion-making. 8. Deployment: + Deploy the trained model into a inputs and make predictions on 9. Monitoring and Maintenance:"* production environment where it can receive new data inouse prices * Monitor the model's performance reguiarly in the live environment. * Retrain the model periodically with new data to keep it updated and accurate. 10. Feedback incegration: * Gather feedback from real estate agents, buyers, ad seliers on the model's predictions. + Use feedback to improve the model's accura Ez Get the Data Working on a machine learning project begins by obtaining the necessary data, which serves as the foundation for all subsequent modeling and analysis. Let us understand the structured approach to effectively gather data for ML projects: 1, Setting Up Your Environment: cy and refine the implementation process * System Preparation: Ensure Python is installed on the system. If not, it can be downloaded and installed from Python's official website.(https://www.python.org/) + Workspace Creation: A dedicated directory for machine learning projects should be created for organizational clarity. This can be set up using Command Prompt: - | ge mkdir C:\ML_Projects C:\> cd C:\ML_ProjectsU7 Data Preparation f : 2 2. Creating an Isolated Environment: * Using Virtual Environments: It is recommended to work in an isolated environment to manage dependencies more effectively and avoid conflicts between projects. The virtual environment tool should be installed and a new environment created: lonp pip a ; ; — joe pip install virtualenv # Install virtualeny | y i 7 | C:\> virtualenv mi_env # Create a new virtual environment named mlenv | C:\> ml_env\scripts\activate # Activate the virtual environment. | While activated, any packages installed using pip will only affect this environment. To exit the virtual environment, simply run: deactivate 3. Installing Necessary Tools: + Essential Libraries: Libraries such as Jupyter, NumPy, pandas, Matplotlib, and Scikit- Learn should be installed if they are not already present. These can be installed using pip, which is Python's package manager. Open Command Prompt and enter the following commands: (ml_env) C:\> python -m pip install --upgrade pip (ml_env) C:\> pip install jupyter matplotlib numpy pandas scikit-learn | 4. Writing Python Scripts: Using a Text Editor or IDE, Create a new file with a py extensicn, for example, data_fetching.py and save it in C:\ML_Projects This scriptisa straightforward tool for data acquisition, automating the process of downloading a dataset and ensuring its availability on the local machine for further data analysis or machine learning tasks. A Python Seript to Get Data from a Repository import os | | import urllib.request ‘https: //raw.githubusercontent. com/ageron/handson-mi/master/datasets/ | DATA_URL housing/housing.csv" : DATA_PATH = os.path.join("C:\\ML_Projects", "datasets", "housing" def fetch_data(data_url=DATA_URL, data_path=DATA PATH): os.makedirs(data_path, exist_ok=True) csv_path = 0s.path.join(data_path, “housing.csv") | urllib.request .urlretrieve(data_url, ,csv_path) print("Data downloaded to:", csv_path) | if _name_ == "_main_"! oe death it th uy