An Improved Heuristic K-Means Clustering Method Using Genetic Algorithm Based Initialization

Abstract

In this paper, we propose methods to remove the drawbacks that commonly afflict the k-means clustering algorithm. We use nature based heuristics to improve the clustering performance offered by the k-means algorithm and also ensure the creation of the requisite number of clusters. The use of GA is found to be adequate in this case to provide a good initialization to the algorithm, and this is followed by a differential evolution based heuristic to ensure that the requisite number of clusters is created without minimal increase in the running time of the algorithm.