Improved Expectation Maximization (EM) Algorithm based on Initial Parameter Selection

Abstract

and high computational cost. Therefore, these remain the need to enhance EM algorithm in its initialization stage to produce good clustering results and to take less computational time. Since good initialization leads to fast convergence, there is a challenge of improving the initialization stage of the EM Algorithm. With that, this study aims to improve the Expectation-Maximization (EM) algorithm based on the initial parameters selection. An improved initial parameters selection technique for EM algorithm is introduced in this study which uses the concept of firefly movement and light intensity of firefly algorithm. A comparison of the clustering performance of the Enhanced EM algorithm (EnEM) to the Standard EM algorithm (StEM) in terms of clustering fitness, clustering error and computing time of the two algorithms is discussed. The application of an enhanced EM algorithm as a clustering method will open more opportunities to discover new knowledge as an outcome of a more precise data analysis.