14
0

Some notes on the kk-means clustering for missing data

Abstract

The classical kk-means clustering requires a complete data matrix without missing entries. As a natural extension of the kk-means clustering for missing data, the kk-POD clustering has been proposed, which ignores the missing entries in the kk-means clustering. This paper shows the inconsistency of the kk-POD clustering even under the missing completely at random mechanism. More specifically, the expected loss of the kk-POD clustering can be represented as the weighted sum of the expected kk-means losses with parts of variables. Thus, the kk-POD clustering converges to the different clustering from the kk-means clustering as the sample size goes to infinity. This result indicates that although the kk-means clustering works well, the kk-POD clustering may fail to capture the hidden cluster structure. On the other hand, for high-dimensional data, the kk-POD clustering could be a suitable choice when the missing rate in each variable is low.

View on arXiv
Comments on this paper