A parallel sampling-PSO-multi-core-K-means algorithm using mapreduce

Clustering is partitioning data into groups, such that data in the same group are similar. Many clustering algorithms are proposed in the literature. K-means is the most used one because of its implementation simplicity and efficiency. Many clustering algorithms are based on the K-means algorithms aiming to improve execution time or clustering quality or both of them. Improving clustering quality can be done by an optimal selection of the initial centroids using for example meta-heuristics. Improving execution time can be performed using parallelism. In this paper, we propose a parallel hybrid K-means based on Google's MapReduce framework for the parallelism and the PSO meta-heuristics for the choice of the initial centroids. This algorithm is used to cluster multi-dimensional data sets. The results proved that using a network of machines to process data improves the execution time and the clustering quality.