MATLAB Implementation of k-means Clustering Algorithm for Binary Classification

This document provides a comprehensive guide to the k-means clustering algorithm implementation in MATLAB. K-means is an unsupervised machine learning algorithm designed to partition datasets into k distinct clusters based on feature similarity. Our implementation specifically focuses on binary classification (two-cluster separation). The complete program package includes both the main driver function and sample datasets, allowing immediate execution and result visualization. The core algorithm operates through these key steps: initial centroid selection using random sampling, Euclidean distance computation between data points and centroids, iterative cluster reassignment based on minimum distance criteria, and centroid recalculation until convergence. For users wishing to modify or extend the program for custom datasets or different cluster counts, understanding the underlying implementation is crucial. The code structure features modular functions for distance calculation (pdist2), centroid initialization (randperm), and convergence checking (norm difference threshold). Key adjustable parameters include maximum iteration limits, convergence tolerance, and centroid initialization methods. We provide sample datasets demonstrating typical clustering scenarios and discuss potential applications in data segmentation, pattern recognition, and preprocessing for classification tasks. The implementation employs vectorized operations for efficient computation and includes visualization components for cluster result plotting. This resource aims to facilitate deeper understanding of k-means clustering mechanics while providing a practical, reusable codebase for MATLAB users working with unsupervised learning algorithms.