Performance Analysis of the EM Algorithm in Conjunction with Algorithms for Determining an Optimal Number of Clusters and Their Centroids, Which Allows Estimating Parameters of Non-Gaussian Interference in Mobile Communication Systems

Abstract

In this article, the “elbow method” and the “average silhouette” algorithms are used for determining an optimal number of clusters. A “shifted” EM algorithm adapted to the number of clusters is put forward, which allows increasing accuracy of the Gaussian mixture distribution approximation that determines the probabilistic structure of the interference distribution. Dependencies of an average number of cycles and the Kullback–Leibler divergence on a number of clusters in an initial mixture are obtained via mathematical modeling with a preset range of parameters. A comparative analysis of the EM algorithm results is carried out.