Hybrid Big Bang-Big Crunch Algorithm for Cluster Analysis

Abstract

Data clustering is an exploratory technique that organizes the data objects into different clusters in a competent way. There are number of techniques reported in clustering field. Several shortcomings associated with these techniques have been identified and resolved such as initial cluster center selection, number of clusters, slow convergence rate, local optima etc. In present work, a hybrid version of the big bang-big crunch (BB-BC) algorithm is developed to optimize clustering problems. The proposed algorithm work in two stages, initialization and optimization. The K-means algorithm act as initiation arbitrator to generate the initial population. While the big bang-big crunch algorithm acts as an optimizer to obtain the best solution. Here, the cluster centers generated from K-means are treated as preliminary population in BB-BC algorithm. The performance of proposed hybrid BB-BC algorithm is examined over seven benchmark datasets and compared with BB-BC, ACO, GA, PSO and K-means clustering algorithms. From the experimental results, it is clarified that proposed algorithm gives better clustering solution than rest of algorithms.