KCK-Means: A Clustering Method Based on Kernel Canonical Correlation Analysis

摘要

Kernel Canonical Correlation Analysis (KCCA) is a technique that can extract common features from a pair of multivariate data, which may assist in mining the ground truth hidden in the data. In this paper, a novel partitioning clustering method called KCK-means is proposed based on KCCA. We also show that KCK-means can not only be run on two-view data sets, but also it performs excellently on single-view data sets. KCK-means can deal with both binary-class and multi-class clustering tasks very well. Experiments with three evaluation metrics are also presented, the results of which reflect the promising performance of KCK-means.