An improved sampling K-means clustering algorithm based on MapReduce

The parallel sampling K-means algorithm can improve the performance and accuracy of clustering, but there are still some problems to be settled, such as samples sensitivity to the initial clustering center and the clustering effect stability. In this paper, an improved parallel sampling K-means clustering algorithm is proposed. First, the sampling from the big data is parallelized in the new algorithm. Second, k′ points are selected randomly from the sample and singular points among them are removed, and k points are obtained for sample clustering. Finally, the matrix of all sample cluster centers is generated and a row of the matrix is selected randomly as initial center to cluster again to get initial cluster center for big data set. The experiment results demonstrate that the improved sampling K-means clustering algorithm has good accuracy and stability for big data clustering.