added 2nd k-means to ensure random samples in a cluster are not close to each other in k-means sample selection