On autonomous k-means clustering

Clustering is a basic tool in unsupervised machine learning and data mining. One of the simplest clustering approaches is the iterative k-means algorithm. The quality of k-means clustering suffers from being confined to run with fixed k rather than being able to dynamically alter the value of k. Moreover, it would be much more elegant if the user did not have to supply the number of clusters for the algorithm. In this paper we consider recently proposed autonomous versions of the k-means algorithm. We demonstrate some of their shortcomings and put forward solutions for their deficiencies. In particular, we examine the problem of automatically determining a good initial candidate as the number of clusters.