Initial Centroid Determination Using Simulated Annealing Algorithm

Abstract

Initial randomly generated centroids are commonly used in k-Means clustering method. Random initial centroids k-Means to be trapped in optimum local solution which results in sub-optimal cluster quality. This study examines Simulated Annealing algorithm in determining initial centroids on k-Means. Each k-Means clustering will be tested on result of reduction and without dimension reduction. Based on the results evaluation of k-Means clustering results with initial centroid Simulated Annealing algorithm improve quality cluster with percentage change value 21.2% in the high dimensional data and 25.1% in the dimension reduction data, this shows that initial centroid calculated Simulated Annealing algorithm is able to obtain the best cluster with significant results.