<?xml version="1.0" encoding="UTF-8"?><rss xmlns:dc="http://purl.org/dc/elements/1.1/" xmlns:content="http://purl.org/rss/1.0/modules/content/" xmlns:atom="http://www.w3.org/2005/Atom" version="2.0" xmlns:cc="http://cyber.law.harvard.edu/rss/creativeCommonsRssModule.html">
    <channel>
        <title><![CDATA[Stories by Karrolla Bhojavva on Medium]]></title>
        <description><![CDATA[Stories by Karrolla Bhojavva on Medium]]></description>
        <link>https://medium.com/@karrollakeerthi277?source=rss-051ba30ac875------2</link>
        <image>
            <url>https://cdn-images-1.medium.com/fit/c/150/150/0*U_S8hk89SSyRXz7a</url>
            <title>Stories by Karrolla Bhojavva on Medium</title>
            <link>https://medium.com/@karrollakeerthi277?source=rss-051ba30ac875------2</link>
        </image>
        <generator>Medium</generator>
        <lastBuildDate>Sat, 21 Nov 2026 22:11:44 GMT</lastBuildDate>
        <atom:link href="https://medium.com/@karrollakeerthi277/feed" rel="self" type="application/rss+xml"/>
        <webMaster><![CDATA[yourfriends@medium.com]]></webMaster>
        <atom:link href="http://medium.superfeedr.com/" rel="hub"/>
        <item>
            <title><![CDATA[A Comparative Study of High-Cardinality Encoding Techniques for Machine Learning]]></title>
            <link>https://medium.com/@karrollakeerthi277/a-comparative-study-of-high-cardinality-encoding-techniques-for-machine-learning-615fe45f7f42?source=rss-051ba30ac875------2</link>
            <guid isPermaLink="false">https://medium.com/p/615fe45f7f42</guid>
            <category><![CDATA[feature-engineering]]></category>
            <category><![CDATA[categorical-encoding]]></category>
            <category><![CDATA[high-cardinality]]></category>
            <category><![CDATA[machine-learning]]></category>
            <category><![CDATA[data-science]]></category>
            <dc:creator><![CDATA[Karrolla Bhojavva]]></dc:creator>
            <pubDate>Wed, 18 Nov 2026 17:41:59 GMT</pubDate>
            <atom:updated>2026-11-18T17:41:59.195Z</atom:updated>
            <content:encoded><![CDATA[<h3>Introduction</h3><figure><img alt="" src="https://cdn-images-1.medium.com/max/1000/0*UHhVnpn6HzQvOrCp" /></figure><p>Categorical encoding is an important step in machine learning preprocessing. Machine learning algorithms generally require numerical input, so categorical values must be converted into numerical representations.</p><p>This becomes challenging when a dataset contains high-cardinality categorical features. High-cardinality features have a large number of unique categories, which can increase the number of features and make model training more complex.</p><p>In this project, I experimented with different categorical encoding techniques and compared their performance using LightGBM and native categorical handling in XGBoost and LightGBM.</p><p>The main objective was to understand how different encoding techniques affect model performance, feature dimensions, and the ability to distinguish between complaint and no-complaint cases.</p><p><strong>Workflow</strong></p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*f8XxtJMuMQXgzeETHrrRtQ.png" /></figure><h3>Why Compare Different Encoding Techniques?</h3><p>Machine learning algorithms cannot directly process categorical features such as <strong>driver IDs</strong>, <strong>pickup ZIP codes</strong>, or <strong>vehicle models</strong>. These features must first be converted into numerical values.</p><p>However, not all encoding techniques preserve information in the same way. Some increase the number of features, while others use statistical information from the target variable. Therefore, selecting an appropriate encoding technique can significantly influence model performance.</p><p>The objective of this project was to compare eight encoding techniques under identical experimental conditions and evaluate their effectiveness using <strong>Accuracy</strong> and <strong>ROC-AUC</strong>.</p><h3>1. Dataset Overview</h3><p>For this project, I used a ride-hailing complaints dataset containing 30,000 records and 10 initial columns.After removing the unnecessary Unnamed: 0 column, I worked with 8 input features and one target variable.</p><p><strong>Target variable:</strong> complained</p><ul><li>0 — No complaint</li><li>1 — Complaint</li></ul><p>The dataset contained the following categorical features:</p><p>Feature-Unique categories</p><p>driver_id-600 , pickup_zip-1500 , dropoff_zip-1500 , vehicle_model-5</p><p>The numerical features were:</p><p>hour , distance_mi , fare , wait_min</p><p>The dataset had an imbalanced target distribution, with approximately 84% no-complaint records and 16% complaint records.</p><h3>2. Train-Test Split</h3><p>I divided the dataset into training and testing sets using an 80:20 split.</p><p>I used stratified sampling to maintain a similar target distribution in both sets.</p><ul><li>Training records: 24,000</li><li>Testing records: 6,000</li></ul><p>I used random_state=42 to make the experiment reproducible.</p><p>To ensure a fair comparison, all encoding techniques were evaluated using the same train-test split, identical model settings, and the same evaluation metrics. The only difference between the experiments was the categorical encoding technique applied.</p><p>Evaluation Metrics<br>• Accuracy<br>• ROC-AUC</p><p>Random State = 42<br>Train-Test Split = 80:20</p><h3>3. Encoding Techniques Implemented</h3><p>I experimented with eight approaches.</p><h3>Experiment 1: Capped One-Hot Encoding</h3><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*TbTqeyU5Xg4lsEaqEQBL9Q.png" /></figure><p>Capped One-Hot Encoding converts each category into binary columns while limiting the number of generated features. Rare categories are grouped together to reduce dimensionality.</p><p><strong>Parameters:</strong><br>OneHotEncoder(max_categories=20, handle_unknown=&#39;infrequent_if_exist&#39;)</p><p><strong>Observation:</strong> The feature count increased from <strong>8 to 69</strong>.</p><h3>Experiment 2: Frequency Encoding</h3><figure><img alt="" src="https://cdn-images-1.medium.com/max/767/1*QivykxRaOquIiXHlUsQtaw.png" /></figure><p>Frequency Encoding replaces each category with its frequency of occurrence in the training data. It preserves the original number of features and is suitable for high-cardinality categorical variables.</p><p><strong>Observation:</strong> The feature count remained <strong>8</strong>.</p><h3>Experiment 3: Target / Mean Encoding</h3><figure><img alt="" src="https://cdn-images-1.medium.com/max/769/1*SaG-P2lArqJoLoPS7AWAXg.png" /></figure><p>Target Encoding replaces each category with the mean value of the target variable for that category. To reduce target leakage, I used <strong>5-fold target encoding</strong> during training.</p><p><strong>Observation:</strong> The feature count remained <strong>8</strong>.</p><h3>Experiment 4: Smoothed Target Encoding</h3><figure><img alt="" src="https://cdn-images-1.medium.com/max/762/1*JKYeRKXiTClXTghtku3W-A.png" /></figure><p>Smoothed Target Encoding combines the category mean with the global target mean, producing more reliable values for categories with few observations.</p><p><strong>Parameters:</strong><br>min_samples_leaf=20, smoothing=10</p><p><strong>Observation:</strong> Smoothing reduced the influence of rare categories.</p><h3>Experiment 5: Leave-One-Out Encoding</h3><figure><img alt="" src="https://cdn-images-1.medium.com/max/1000/0*6FKnfcYgQ5KD_sge" /></figure><p>Leave-One-Out Encoding calculates the target mean for each category while excluding the current training observation. This helps reduce target leakage and improve model generalization.</p><p><strong>Parameters:</strong><br>sigma=0.05, random_state=42</p><p><strong>Observation:</strong> Achieved the highest <strong>ROC-AUC (0.7147)</strong> with an <strong>Accuracy of 0.6763</strong>.</p><h3>Experiment 6: Weight of Evidence (WoE) Encoding</h3><figure><img alt="" src="https://cdn-images-1.medium.com/max/769/1*WOTFeAQ2GzvijLWzXYRaSA.png" /></figure><p>Weight of Evidence (WoE) Encoding represents the relationship between each category and the target classes. It is commonly used for binary classification problems.</p><p><strong>Observation:</strong> Achieved a <strong>ROC-AUC of 0.7140</strong>.</p><h3>Experiment 7: XGBoost Native Categorical Handling</h3><figure><img alt="" src="https://cdn-images-1.medium.com/max/762/1*D-8mGqloKOjVZBZftbNELw.png" /></figure><p>XGBoost can process categorical features directly without manual encoding by using native categorical support.</p><p><strong>Parameters:</strong><br>enable_categorical=True, tree_method=&#39;hist&#39;</p><p><strong>Observation:</strong> Achieved an <strong>Accuracy of 0.7518</strong> and a <strong>ROC-AUC of 0.6884</strong>.</p><h3>Experiment 8: LightGBM Native Categorical Handling</h3><figure><img alt="" src="https://cdn-images-1.medium.com/max/759/1*ttQ7EAL9kUW7kDjZWr6s3g.png" /></figure><p>LightGBM natively handles categorical features by learning optimal category splits during training, eliminating the need for manual encoding.</p><p><strong>Observation:</strong> Achieved the highest <strong>Accuracy (0.7608)</strong> among all experiments.</p><h3>4. Model Evaluation Metrics</h3><p>The models were evaluated using <strong>Accuracy</strong> and <strong>ROC-AUC</strong>. Accuracy measures the overall percentage of correctly classified instances, while ROC-AUC evaluates the model’s ability to distinguish between complaint and no-complaint cases across different classification thresholds. Since the dataset was imbalanced, ROC-AUC was considered alongside Accuracy to provide a more reliable assessment of model performance.</p><h3>5. Experimental Results</h3><p>After evaluating all eight encoding techniques under the same experimental setup, I compared their performance using <strong>Accuracy</strong> and <strong>ROC-AUC</strong>. The results are summarized in the table below.</p><p>Encoding TechniqueAccuracyROC-AUC-Capped One-Hot(0.62380.5471), Frequency(0.63400.5611), Target Encoding(0.66570.7083), Smoothed Target(0.67580.7126), Leave-One-Out(0.6763<strong>0.7147), </strong>WoE(0.65980.7140), XGBoost Native(0.75180.6884), LightGBM Native(<strong>0.7608</strong>0.6929).</p><h3>Key observations</h3><ol><li>Leave-One-Out Encoding achieved the highest ROC-AUC score of 0.7147.</li><li>WoE achieved a very similar ROC-AUC score of 0.7140.</li><li>Native LightGBM achieved the highest Accuracy of 0.7608.</li><li>Smoothed Target Encoding performed better than normal Target Encoding in both Accuracy and ROC-AUC.</li><li>Capped One-Hot and Frequency Encoding had lower ROC-AUC scores than the target-based encoding techniques in these experiments.</li><li>Native categorical handling produced higher Accuracy than the manual encoding approaches in this experiment.</li></ol><h3>6. Why Did Leave-One-Out and WoE Perform Well?</h3><p>Leave-One-Out Encoding uses target information while excluding the current training observation’s target value. This helps reduce self-target leakage.</p><p>WoE represents the relationship between categorical values and the positive and negative target classes.</p><p>Both techniques achieved strong ROC-AUC results in this experiment.</p><p>However, their performance differences were small. Leave-One-Out achieved 0.7147, while WoE achieved 0.7140.</p><p>This shows that different encoding techniques can preserve useful information in different ways.</p><h3>7. Important Observation About Class Imbalance</h3><p>Although Native LightGBM achieved 76.08% Accuracy, the classification report showed that its recall for the complaint class was 0.36.</p><p>This means the model identified approximately 36% of actual complaint cases at the default classification threshold.</p><p>Therefore, Accuracy alone does not tell the complete story.</p><p>For a complaint prediction problem, identifying complaint cases may be important. Precision, recall, F1-score, and ROC-AUC should also be considered when selecting a model.</p><h3>8. Conclusion</h3><p>This comparative study demonstrates that the choice of categorical encoding technique can significantly influence machine learning performance when working with high-cardinality categorical features.</p><p>Among the evaluated methods, <strong>Leave-One-Out Encoding</strong> achieved the highest <strong>ROC-AUC (0.7147)</strong>, while <strong>Native LightGBM</strong> achieved the highest <strong>Accuracy (0.7608)</strong>. Overall, target-based encoding techniques outperformed simple encoding methods in terms of ROC-AUC.</p><p>The results highlight that there is no single best encoding technique for every dataset. Selecting an appropriate encoding method should depend on the dataset characteristics, the machine learning model, and the evaluation metric.</p><p><strong>Tools and Libraries Used</strong></p><ul><li>Python</li><li>Pandas</li><li>NumPy</li><li>Scikit-learn</li><li>Category Encoders</li><li>LightGBM</li><li>XGBoost</li><li>Matplotlib</li><li>Seaborn</li></ul><p><strong>Key Learning:</strong> The best encoding technique should be selected based on the evaluation metric and the actual requirements of the problem, rather than Accuracy alone.</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=615fe45f7f42" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Understanding Machine Learning: From Basics to Real-World Applications]]></title>
            <link>https://medium.com/@karrollakeerthi277/understanding-machine-learning-from-basics-to-real-world-applications-c253ea6438c2?source=rss-051ba30ac875------2</link>
            <guid isPermaLink="false">https://medium.com/p/c253ea6438c2</guid>
            <dc:creator><![CDATA[Karrolla Bhojavva]]></dc:creator>
            <pubDate>Fri, 30 Oct 2026 11:34:24 GMT</pubDate>
            <atom:updated>2026-10-30T11:34:24.758Z</atom:updated>
            <content:encoded><![CDATA[<figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*1ZTN-AlgqFPfithAa0Syig.jpeg" /><figcaption>LEARN- ADAPT- PREDICT</figcaption></figure><h3>Introduction</h3><p>Machine Learning (ML) is a branch of <strong>Artificial Intelligence (AI)</strong> that allows computers to <strong>learn from data</strong> and <strong>improve their performance over time</strong> without being specifically programmed for every task</p><p>Instead of writing detailed instructions for every possible situation, we give the computer <strong>data</strong>. The computer studies the data, finds patterns, and uses those patterns to make predictions or decisions on new data. Today, machine learning is widely used in healthcare, banking, education, e-commerce, agriculture, and many other industries.</p><p>To understand how machine learning works, we need to explore its evolution, learning methods, different types of models, and real-world applications.</p><h3>The Evolution of Machine Learning</h3><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/0*ZJ3Vu255Y0zRMAUb" /><figcaption>RULE BASED SYSTEM-MACHINE LEARNING</figcaption></figure><h4>1. Rule-Based Programming and Its Limitations</h4><p>Rule-based programming is a traditional approach in which developers write specific instructions and conditions for computers to follow.</p><p>For example, a banking system might use the following rule:</p><ul><li>If the account balance is greater than ₹5,000, approve the transaction.</li><li>Otherwise, reject the transaction.</li></ul><p><strong>Limitations:</strong></p><ul><li>Every rule must be manually created.</li><li>It cannot easily handle complex situations.</li><li>It struggles with unexpected patterns.</li><li>Updating rules becomes difficult when data changes.</li></ul><h4>2. The Evolution and Need for Machine Learning</h4><p>As technology developed, organizations started generating huge amounts of data. Traditional programming became insufficient for identifying complex patterns and making predictions.</p><p>Machine Learning emerged as a solution that allows computers to learn patterns from historical data and make decisions or predictions.</p><p>For example, instead of manually writing rules to identify spam emails, we can train an ML model using examples of spam and non-spam emails.</p><p><strong>Why do we need Machine Learning?</strong></p><ul><li>To analyze large amounts of data.</li><li>To identify hidden patterns.</li><li>To automate repetitive tasks.</li><li>To make predictions based on historical information.</li><li>To improve performance with experience and new data.</li></ul><h4>3. The Role of Data Analysis in Building Machine Learning Models</h4><p>Data analysis plays an important role in machine learning because models learn from data. Poor-quality data can lead to inaccurate predictions.</p><p>Before training a model, data analysts and data scientists perform several activities:</p><ol><li>Data Collection — Gathering information from different sources.</li><li>Data Cleaning — Handling missing values, duplicates, and incorrect data.</li><li>Exploratory Data Analysis (EDA) — Understanding patterns, relationships, and distributions.</li><li>Feature Selection — Identifying the most useful variables.</li><li>Data Preparation — Transforming data into a suitable format for model training.</li></ol><p><strong>Example:</strong> In a house price prediction project, we analyze house size, location, number of bedrooms, and historical prices to understand which factors influence property prices.</p><h3>Understanding How ML Models Learn</h3><figure><img alt="" src="https://cdn-images-1.medium.com/max/887/0*fZjaiO0hWPmNqHTc" /></figure><h4>1. Parametric and Non-Parametric Models</h4><p>Machine learning models can be categorized based on how they learn patterns from data.</p><p><strong>Parametric Models</strong></p><p>Parametric models assume a particular mathematical structure and learn a fixed number of parameters from training data.</p><p>Examples:</p><ul><li>Linear Regression</li><li>Logistic Regression</li><li>Naive Bayes</li></ul><p>Example: Linear Regression learns coefficients to predict house prices based on house size.</p><p><strong>Non-Parametric Models</strong></p><p>Non-parametric models do not assume a fixed mathematical structure. Their complexity can grow with the amount of training data.</p><p>Examples:</p><ul><li>K-Nearest Neighbors (KNN)</li><li>Decision Trees</li><li>Random Forests</li></ul><p>Example: A KNN model predicts whether a customer will purchase a product by comparing the customer with similar existing customers.</p><h4>2. Generalization and Memorization</h4><p><strong>Memorization</strong></p><p>Memorization happens when a machine learning model learns the training examples too closely instead of identifying general patterns.</p><p>This can lead to overfitting, where the model performs well on training data but poorly on unseen data.</p><p><strong>Generalization</strong></p><p>Generalization is the ability of a model to make accurate predictions on new, unseen data.</p><p>A good ML model should learn meaningful patterns rather than simply memorize the training examples.</p><p>Example:</p><p>Imagine a student who memorizes every answer in a textbook. The student may perform well on familiar questions but struggle with new questions.</p><p>A student who understands the concepts can answer both familiar and unfamiliar questions.</p><p>Similarly, a machine learning model should understand patterns and generalize to new data.</p><h4>3. The Rat-and-Food and Pigeon-Superstition Examples</h4><p>These examples help explain how learning can happen through experience, rewards, and associations.</p><p><strong>Rat-and-Food Example</strong></p><p>Imagine a rat placed inside a maze with food at a particular location.</p><ul><li>Initially, the rat explores different paths.</li><li>When it finds food, it receives a reward.</li><li>Through repeated experiences, it learns which path leads to food.</li><li>Over time, it becomes better at finding the food.</li></ul><p>This illustrates learning through trial and error, which is closely related to reinforcement learning.</p><p><strong>Pigeon-Superstition Example</strong></p><p>Psychologist B. F. Skinner studied superstitious behaviour in pigeons.</p><p>In an experiment, pigeons received food at regular intervals, regardless of what they were doing. Some pigeons developed repeated behaviours, such as turning or moving their heads, as if those actions caused the food reward.</p><p>This demonstrates how an organism can develop an incorrect association between an action and a reward.</p><p><strong>Key Learning:</strong> A model should identify genuine relationships in data rather than learning misleading patterns or accidental associations.</p><h3>Types of Machine Learning</h3><p>Machine Learning is generally divided into four major approaches.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*z98zw2yw-TcgYd6QLiRBwg.png" /><figcaption>TYPES OF MACHINE LEARNING</figcaption></figure><h3>1. Supervised Learning</h3><p>Supervised learning is a machine learning approach in which a model learns from labelled data.</p><p>The training dataset contains both input features and the correct output.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/568/0*lSTuzOOLql8vthAz" /></figure><p><strong>How it works:</strong></p><ol><li>Collect labelled data.</li><li>Provide input and correct output to the model.</li><li>The model learns relationships between them.</li><li>Test the model using unseen data.</li><li>Use the trained model to make predictions.</li></ol><p><strong>Real-world example:</strong> Email spam detection.</p><p>The model learns from emails labelled as spam or not spam and predicts the category of new emails.</p><h3>2. Unsupervised Learning</h3><p>Unsupervised learning is an approach in which a model learns from unlabelled data.</p><p>It identifies hidden patterns, similarities, and groups without predefined output labels.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/0*EjTSz3WCOeObrnME" /></figure><p><strong>How it works:</strong></p><ol><li>Collect unlabelled data.</li><li>Provide data to the model.</li><li>The model identifies patterns or similarities.</li><li>It creates groups or discovers relationships.</li></ol><p><strong>Real-world example:</strong> Customer segmentation.</p><p>An e-commerce company groups customers based on purchasing behaviour, spending habits, and interests.</p><h3>3. Reinforcement Learning</h3><p>Reinforcement learning is an approach in which an agent learns by interacting with an environment and receiving rewards or penalties.</p><figure><img alt="" src="https://cdn-images-1.medium.com/proxy/0*NusJvNZPfoLyz-Yv" /></figure><p><strong>How it works:</strong></p><ol><li>An agent observes the environment.</li><li>It takes an action.</li><li>The environment provides a reward or penalty.</li><li>The agent learns from the result.</li><li>It improves its future decisions.</li></ol><p><strong>Real-world example:</strong> Self-driving cars.</p><p>A learning system can use feedback from driving simulations to improve decisions such as braking, turning, and avoiding obstacles.</p><h3>4. Self-Supervised Learning</h3><p>Self-supervised learning is an approach in which a model creates learning signals from the data itself instead of depending on manually labelled examples.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/0*QRowVJ_v6TGP4_ki" /></figure><p><strong>How it works:</strong></p><ol><li>Collect large amounts of raw data.</li><li>Create a learning task from the data.</li><li>Train the model to predict missing or hidden information.</li><li>Learn useful patterns and representations.</li><li>Apply the learned knowledge to other tasks.</li></ol><p><strong>Real-world example:</strong> Language models.</p><p>A language model learns by predicting missing or next words in sentences using large amounts of text.</p><h3>5. Semi-Supervised Learning</h3><p>Semi-Supervised Learning is an approach in which a model learns from a small amount of labelled data and a large amount of unlabelled data.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/0*E9ZECttXyJRgnmdm" /></figure><p>How it works:</p><ol><li>Collect a small amount of labelled data.</li><li>Collect a large amount of unlabelled data.</li><li>Train the model using labelled examples.</li><li>Identify patterns in unlabelled data.</li><li>Improve the model’s predictions using the learned information.</li></ol><p>Real-world example: Image Classification.</p><p>A model learns to identify cats, dogs, and birds using a few labelled images and a large collection of unlabelled images.</p><h3>Types of Machine Learning Problems</h3><p>Machine learning problems are commonly divided into two important categories: Classification and Regression.</p><h4>1. Classification</h4><p>Classification is a supervised learning problem in which the model predicts a category or class label.</p><p><strong>Examples:</strong></p><ul><li>Email: Spam or Not Spam.</li><li>Banking: Loan Approved or Rejected.</li><li>Healthcare: Disease Positive or Negative.</li><li>Customer Analysis: Likely to Leave or Likely to Stay.</li></ul><p><strong>How it works:</strong></p><p>The model learns from labelled examples and assigns a category to new observations.</p><h4>2. Regression</h4><p>Regression is a supervised learning problem in which the model predicts a continuous numerical value.</p><p><strong>Examples:</strong></p><ul><li>Predicting house prices.</li><li>Estimating monthly sales.</li><li>Forecasting temperature.</li><li>Predicting customer spending.</li></ul><p><strong>How it works:</strong></p><p>The model learns the relationship between input variables and numerical outputs to estimate values for new observations.</p><h4>Classification vs Regression</h4><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/0*H6FIFC-SKVUF5JdC" /></figure><p>Classification is a supervised machine learning technique used to predict categories or labels. It helps determine which group an observation belongs to. For example, an email can be classified as Spam or Not Spam, and a bank can classify a loan application as Approved or Rejected. Common classification algorithms include Logistic Regression, Decision Tree, and Random Forest Classifier. Classification models are evaluated using metrics such as Accuracy, Precision, Recall, and F1-Score.</p><p>Regression is a supervised machine learning technique used to predict continuous numerical values. It helps estimate quantities based on relationships between input variables. For example, a regression model can predict house prices based on location, size, and number of bedrooms or estimate future sales based on historical data. Common regression algorithms include Linear Regression, Decision Tree Regressor, and Random Forest Regressor. Regression models are evaluated using MAE, MSE, RMSE, and R².</p><h3>Machine Learning in the Real World</h3><p>Machine Learning has become an important technology across different industries.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*DZOG3H20Y_0rWJBdqE8-vg.png" /></figure><h4>1. Healthcare</h4><p>Machine Learning helps healthcare professionals analyze medical information and support clinical decisions.</p><p>Applications:</p><ul><li>Medical image analysis.</li><li>Disease risk prediction.</li><li>Patient monitoring.</li><li>Drug discovery.</li></ul><p>Example: ML models can help identify suspicious patterns in medical images, supporting doctors in diagnosis.</p><h3>2. Finance and Banking</h3><p>Financial institutions use machine learning to improve security and analyze financial activities.</p><p>Applications:</p><ul><li>Fraud detection.</li><li>Credit risk assessment.</li><li>Loan approval support.</li><li>Financial forecasting.</li></ul><p>Example: Banks use ML models to identify unusual transaction patterns that may indicate fraudulent activity.</p><h4>3. E-Commerce</h4><p>Online shopping platforms use machine learning to understand customer behaviour and improve shopping experiences.</p><p>Applications:</p><ul><li>Product recommendations.</li><li>Customer segmentation.</li><li>Demand forecasting.</li><li>Personalized offers.</li></ul><p>Example: An online shopping platform recommends products based on customers’ previous purchases and browsing behaviour.</p><h4>4. Education</h4><p>Machine Learning helps educational institutions understand student performance and provide personalized learning experiences.</p><p>Applications:</p><ul><li>Student performance prediction.</li><li>Personalized learning recommendations.</li><li>Automated assessment support.</li><li>Identifying students who may need additional support.</li></ul><p>Example: An educational platform recommends learning materials based on a student’s performance and progress.</p><h4>5. Agriculture</h4><p>Machine Learning supports farmers in improving agricultural productivity and resource management.</p><p>Applications:</p><ul><li>Crop disease detection.</li><li>Weather and yield prediction.</li><li>Soil analysis.</li><li>Smart irrigation.</li></ul><p>Example: ML models can analyze crop images to identify signs of plant diseases at an early stage.</p><h4>6. Recommendation Systems</h4><p>Recommendation systems use machine learning to suggest products, movies, music, and other content based on user preferences and behaviour.</p><p>Applications:</p><ul><li>Movie recommendations.</li><li>Music suggestions.</li><li>Product recommendations.</li><li>Personalized social media content.</li></ul><p>Example: Netflix recommends movies and shows based on viewing history and similar users’ preferences.</p><h3>Conclusion</h3><p>Machine Learning is transforming the way people and organizations use data to solve real-world problems.</p><p>Understanding its evolution, learning approaches, model behaviour, and problem types provides a strong foundation for studying advanced machine learning concepts.</p><p>Data analysis is an essential part of this journey because clean, meaningful, and well-prepared data helps build more reliable models.</p><p>As technology continues to develop, machine learning will play an increasingly important role in creating intelligent solutions across different industries.</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=c253ea6438c2" width="1" height="1" alt="">]]></content:encoded>
        </item>
    </channel>
</rss>