A description length approach to determining the number of k-means clusters

Hiromitsu Mizutani, Ryota Kanai

We present an asymptotic criterion to determine the optimal number of clusters in k-means. We consider k-means as data compression, and propose to adopt the number of clusters that…