MFWK-Means: Minkowski metric Fuzzy Weighted K-Means for high dimensional data clustering

Abstract

This paper presents a clustering algorithm, namely MFWK-Means, which is a novel extension of K-Means clustering to the case of fuzzy clusters and weighted features. First, the Weighted K-Means criterion utilizing Minkowski metric is adopted to solve the problem of feature selection for high dimensional data. Then, a further extension to the case of fuzzy clustering is presented to group datasets with natural fuzziness of cluster boundaries. Also, we adopt an intelligent version of K-Means, using Mirkin's method of Anomalous Pattern for initialization. Our new Minkowski metric Fuzzy Weighted K-Means (MFWK-Means) is experimentally validated on both benchmark datasets and synthetic datasets. MFWK-Means is shown to be competitive and more stable against noise in comparison with a variety of versions of K-Means based methods. Moreover, in most situations it reaches the highest clustering accuracy at wider intervals of Minkowski exponent.