Introduction to Machine Learning

Birla Institute Of Technology And Science–Pilani (BITS–Pilani) via Coursera

Go to class Write review

Details

Go to class

Provider

Coursera
Pricing

Paid Course
Languages

English
Certificate

Certificate Available
Effort

10 weeks, 5 hours a week
Sessions

Self-Paced
Level

Intermediate
Subtitles

English

Overview

Google, IBM & Meta Certificates – 40% Off

One plan covers every Professional Certificate on Coursera.

Unlock All Certificates

This course offers a comprehensive introduction to the principles and techniques of machine learning, focusing on building algorithms that can learn patterns from data to make predictions and informed decisions. It encompasses various kinds of learning, primarily categorized into supervised and unsupervised learning. In supervised learning, models are trained on labeled data for tasks such as regression and classification, while unsupervised learning deals with discovering hidden structures in unlabeled data, such as in clustering. Through practical and theoretical modules, learners will gain a deep understanding of model selection, including the use of cross-validation and evaluation metrics such as accuracy, precision, recall, and F1 score. The course also introduces the Bayesian learning framework, offering a probabilistic perspective on learning through concepts like prior and posterior distributions, Maximum Likelihood (ML) estimation, Maximum a Posteriori (MAP) hypothesis, and the Minimum Description Length (MDL) principle. Key concepts such as the bias-variance tradeoff and regularisation techniques are explored to help students develop models that generalise well to unseen data. Learners will also work with foundational models such as Naive Bayes classifiers, both Gaussian and Multinomial, and linear models like Linear Regression and Logistic Regression, supported by methods like feature scaling and gradient descent. The course further extends to non-linear models using polynomial transformations, as well as decision trees built using criteria like entropy and Gini index, with an emphasis on pruning to control overfitting. Students will also explore instance-based learning methods such as k-Nearest Neighbours (KNN) and advanced classification techniques like Support Vector Machines (SVMs), including the use of kernel functions for handling complex, non-linear data. By the end of the course, students will be equipped with a solid foundation in machine learning theory and practice, preparing them for further study or real-world applications in data science and AI.

Syllabus

Introduction

This module provides an overview of machine learning, its scope, and real-world applications. It introduces various types of learning. This module introduces the foundational concepts of Machine Learning, guiding learners through its definition, scope, types, and core workflow. It begins with a broad understanding of what Machine Learning is, its real-world applications across domains, and how it relates to and differs from traditional programming and Artificial Intelligence. The second lesson dives into the taxonomy of Machine Learning, exploring the four main paradigms: supervised, unsupervised, reinforcement, and semi-supervised learning. In the third lesson, learners are taken through the standard workflow of building a Machine Learning model — from problem formulation to deployment, including revised, unsupervised, semi-supervised, and reinforcement learning—and outlines the typical workflow of building ML models.

Linear Models of Regression - 1

This module introduces linear regression as a foundational supervised learning algorithm for predicting continuous outcomes. Learners will explore both analytical (closed-form) and iterative (gradient descent) approaches to model learning. Key concepts such as hyperparameter tuning, regularisation, and model evaluation are explained. The module also addresses the critical concepts of underfitting and overfitting and introduces strategies to mitigate these issues.

Linear Models of Regression - 2

This module discusses linear regression models and the practical challenges that arise when training them. You will explore techniques to mitigate overfitting and underfitting, improve model performance, and enhance generalisation. Through short video lessons, hands-on quizzes, and selected readings, you will learn key concepts such as regularisation, bias–variance tradeoff, feature engineering, feature normalisation, and polynomial transformations. The module also covers hyperparameter tuning and its role in optimising regression models. By the end, you will be equipped with a strong foundation to build, evaluate, and refine linear regression models in real-world scenarios.

Logistic Regression

This module introduces logistic regression, a fundamental classification algorithm used to predict categorical outcomes, particularly binary classes. Learners will explore the mathematical formulation of logistic regression using the sigmoid function and understand how the log-likelihood function is optimised using gradient descent. The module covers key aspects of model learning, including the role of loss functions (cross-entropy), hyperparameter tuning (learning rate, regularisation), and decision thresholds. Core challenges like overfitting and underfitting are addressed, along with regularisation strategies (L1, L2) to control model complexity. Emphasis is placed on evaluating model performance using metrics.

Bayesian Learning - 1

This module introduces the foundational principles of probabilistic learning used in supervised machine learning. Students will explore two fundamental parameter estimation techniques: Maximum Likelihood Estimation (MLE) and Maximum A Posteriori (MAP) estimation. The module covers how these methods infer model parameters from data, with and without prior knowledge. Building on these concepts, the module covers the Naive Bayes Classifier, a simple yet powerful probabilistic model that uses Bayes’ Theorem under the assumption of feature independence. Students will learn how MLE and MAP influence the training of Naive Bayes models and will understand the need for smoothing techniques to handle zero-frequency problems in categorical data. Students will also gain hands-on experience in applying them to real-world classification problems.

Bayesian Learning - 2

This module introduces the foundational principles of probabilistic learning using the Naive Bayes classifier in supervised machine learning. Learners will explore core probabilistic concepts such as conditional independence, product and chain rules, and Bayes theorem. The module illustrates how the Naive Bayes classifier can be applied to classification problems. It also addresses practical challenges such as the zero-probability problem and introduces smoothing techniques like Laplacian smoothing. Additionally, learners will compare generative and discriminative models and gain hands-on experience implementing Naive Bayes using Python and scikit-learn.

Decision Trees

This module introduces Decision Trees, a versatile and interpretable model used for both classification and regression tasks. Learners will explore the core principles of tree construction, including how decisions are made based on feature splits using measures such as information gain, gain ratio and gini index. The module explains how the choice of splitting criterion affects tree structure and performance. It also covers the issues of overfitting and underfitting, and introduces strategies to mitigate them using pre-pruning and post-pruning. Students will also gain hands-on experience in applying them to real-world classification problems.

Support Vector Machine

This module introduces the fundamental concepts and practical applications of Support Vector Machines (SVM), a powerful supervised learning algorithm used for classification and regression tasks. Students will explore the theoretical foundations of SVM, including the idea of maximising the margin between data classes, kernel methods for handling non-linearly separable data, and optimisation techniques used to train SVM models. The module will cover both linear and non-linear SVMs, soft margin classification to handle noisy data, and multi-class extensions. Practical sessions will include the implementation of SVMs using popular machine learning libraries and real-world case studies. Students will also gain hands-on experience in applying them to real-world classification problems.

Support Vector Machine - 2

This module provides a comprehensive introduction to Support Vector Machines (SVMs), one of the most powerful and widely used algorithms in supervised learning. Learners will explore the limitations of hard-margin SVMs and understand how soft margins, slack variables, and hinge loss enable SVMs to handle real-world, imperfectly separable data. The module then progresses to non-linear SVMs, introducing feature transformations and the kernel trick for high-dimensional decision boundaries. Key properties, practical limitations, and comparisons with logistic regression are also discussed. Finally, learners will implement SVMs in Python using scikit-learn to reinforce conceptual understanding with hands-on practice.

K-Nearest Neighbour (KNN)

This module introduces the k-Nearest Neighbours (k-NN) algorithm, a simple yet powerful non-parametric technique used for both classification and regression tasks in supervised learning. The module covers the theoretical foundations of k-NN, including distance metrics, the importance of selecting an appropriate k, and how the algorithm makes predictions based on the similarity between data points. Learners will explore practical aspects such as feature scaling, the impact of dimensionality. Through hands-on coding exercises and real-world datasets, this module equips students with the skills to implement k-NN using Python and to critically assess its performance relative to other algorithms.