Introduction to Machine Learning

• Machine Learning (ML) is an AI subfield enabling systems to learn from data and enhance task performance without explicit programming by building models based on training data to make informed predictions or decisions.

Types of Machine Learning

• Supervised Learning uses labeled data (input-output pairs) to train models that map inputs to known outputs.
• Supervised learning is used for classification to predict discrete labels like spam detection and image recognition.
• Supervised learning is also used for regression to predict continuous values like housing prices or temperature forecasting.
• Algorithms used in supervised learning include Linear Regression, Decision Trees, Support Vector Machines (SVM), and Neural Networks.
• Supervised learning can be used to predict student grades based on study hours and past performance.
• Unsupervised Learning identifies patterns in unlabeled data without predefined outputs.
• Unsupervised learning is leveraged for clustering to group similar data points, such as customer segmentation.
• Unsupervised learning does dimensionality reduction, simplifying data while retaining key features, such as PCA for visualization.
• Unsupervised learning does anomaly detection, and identifies outliers, like fraud detection.
• Algorithms for unsupervised learning are like K-means, Hierarchical Clustering, Autoencoders, and t-SNE.
• Unsupervised learning groups news articles by topics without prior labels.
• Reinforcement Learning (RL) allows an agent to learn by interacting with an environment to maximize cumulative rewards (points, penalties).
• Game-playing AI (AlphaGo), robotics, and autonomous driving use reinforcement learning.
• Key concepts in reinforcement learning are exploration vs. exploitation and policy optimization.
• Common algorithms used in reinforcement learning include Q-Learning, Deep Q-Networks (DQN), and Proximal Policy Optimization (PPO).
• Reinforcement learning is exemplified by a robot learning to walk via reward for forward movement.
• Semi-Supervised Learning combines small labeled data with large unlabeled data, useful when labeling is costly.
• Medical image analysis, where only a few scans are labeled, utilizes semi-supervised learning.

Linear Regression

• Linear regression predicts continuous numerical values based on one or more input variables.
• It establishes a linear relationship between the dependent (target) and independent variables (features).

Types of Linear Regression

• Simple Linear Regression models the relationship between one independent variable (X) and one dependent variable (Y) using a linear equation.
• In simple linear regression, the equation is Y = B0 + B1X + ε.
• Y represents the dependent variable or the target we aim to predict.
• X stands for the independent variable or the feature influencing Y.
• B0 is the intercept, also known as the bias term.
• B1 is the coefficient, representing the slope of the line.
• ε signifies the error term or the residual, accounting for the difference between actual and predicted values.
• Multiple Linear Regression extends beyond simple linear regression to incorporate multiple independent variables for prediction.
• Multiple linear regression equation: Y = B0 + B1X1 + B2X2 + ... + BnXn + ε, where X1, X2, ..., Xn are distinct predictor variables.
• The general form of a line is Y = B0 + B1X + ε.
• B0 is the intercept, indicating where the line starts.
• B1 is the slope, showing how much Y changes for each unit increase in X.
• E is the error, representing the difference between actual and predicted values.

Overfitting and Underfitting in Machine Learning

• Overfitting and underfitting are common problems in machine learning models, including linear regression, that affect how well a model generalizes to unseen data.
• Machine Learning: the model adjusts the line to minimize the differences between actual and predicted values.
• The least squares method minimizes the Mean Squared Error (MSE).
• Error minimization to predict new values of Y for given X values once the best-fit line is found.
• Overfitting occurs when a model learns not only the underlying pattern in the training data but also the noise, performing very well on training data but poorly on test data.
• Underfitting occurs when a model is too simple and fails to capture the underlying trend in the data, performing poorly on both training and test data.

Types of Error Functions Used in Linear Regression

• In linear regression, error functions (also called loss functions) measure how well the predicted values match the actual values, and the goal is to minimize the error for accurate predictions.
• Common error functions include Mean Squared Error (MSE), Root Mean Squared Error (RMSE), Mean Absolute Error (MAE), Huber Loss, and Log-Cosh Loss.