Random centroid initialization for improving centroid-based clustering

A method for improving centroid-based clustering is suggested. The improvement built on diversification of the k-means++ initialization. algorithm claimed to be a better version k-means tested by computational set-up, where dataset size, number features, and clusters are varied. statistics obtained testing have shown that, in roughly 50 % instances cluster, outputs worse results than with random centroid impact initialization solidifies as both size features increase. In order reduce possible underperformance k-means++, run separate processor core parallel running algorithm, whereupon result selected. runs set not less that k-means. By incorporating seeding initialization, gains about 0.05 accuracy every second instance cluster.