Performance Comparison of K-Means and Expectation Maximization with Gaussian Mixture Models for Clustering EE6540 Final Project

In this article, we explore the theoretical aspects of the expectation maximization algorithm and how it can be applied to estimation of data as a Gaussian mixture model. We form comparisons and show how by the simplification of some parameters we can form the heuristic k-means algorithm. We then demonstrate through the authors’ code several situations where the EM-GMM algorithm performs significantly better than k-means and touch upon potential implications of these findings for arbitrary distribution approximation. Finally, we form some conclusions about scenarios where EM-GMM should be used over k-means and how to use them for classification.