Multi-Class Data Clustering with K-means, GMM, and Hierarchical Clustering Algorithms

In this article, we implement multi-class data clustering using three prominent algorithms: K-means, Gaussian Mixture Models (GMM), and hierarchical clustering. We conduct a detailed performance analysis of each algorithm and provide a comprehensive experimental report that documents our methodology, dataset characteristics, and experimental outcomes. The K-means implementation involves centroid initialization optimization and iterative distance minimization using Euclidean metrics. Our GMM approach employs Expectation-Maximization (EM) algorithm for parameter estimation with covariance matrix regularization. The hierarchical clustering implementation includes both agglomerative and divisive methods with linkage criterion optimization. We discuss the comparative advantages and limitations of each algorithm, including K-means' computational efficiency versus GMM's probabilistic framework and hierarchical clustering's dendrogram visualization capabilities. Furthermore, we propose enhancement directions such as automated cluster number selection using silhouette analysis and kernel-based similarity measures for handling non-linear data structures. Our research provides significant references for multi-class data clustering applications and offers valuable insights for future investigations in unsupervised learning domain.