Introduction to data analytics and information visualization; methods and metrics for validation; bias-variance trade- off; data visualization and understanding; data preprocessing; supervised learning (classification and regression); unsupervised learning; association rule mining; feature subset selection; metaheuristics; PCA; ANN and Multilayer Perceptron.

This course develops data analytics as a sequence of optimization problems. Nearly every method in this course is defined by three choices: (i) the loss that is minimized, (ii) the constraint or regularizer that restricts the solution, and (iii) whether the resulting problem can be solved to optimality. The course is organized so that these three choices are filled in for every technique introduced. Supervised learning, unsupervised learning, association rule mining, and feature subset selection are covered within this single frame.

The toolbox includes validation and cost-sensitive evaluation; decision trees and information-theoretic splitting criteria; Naïve Bayes and instance-based learning; the bias-variance decomposition; multiple linear regression with ridge, lasso, and logistic extensions; convexity, Lagrangian duality, and the KKT conditions; support vector machines and support vector regression; partitional, hierarchical, probabilistic, and density-based clustering; the Apriori algorithm; outlier detection, missing data, imbalance, and leakage; principal component analysis; filter, wrapper, and embedded methods for feature subset selection; metaheuristics such as simulated annealing, tabu search, and genetic algorithms; and ensemble methods. Each technique is developed from a worked numerical example and then derived in depth. Note that no software is required for this course, however all approaches can easily be verified by open source code snippets. Coding has traditionally been at the heart of what is taught in this course. However, given recent advances in AI-assisted coding, the actual aim is for students to understand each method rigorously: (i) carry it out by hand, (ii) state the assumptions it requires, and (iii) explain why it works - rather than simply calling a library function.

Lecture notes.

None. A calculator is sufficient.