A Novel Fuzzy Kernel Clustering Algorithm for Outlier Detection

Abstract

Outlier detection is an integral part of data mining and has attracted much attention recently. It may be that an outlier implies the most important feature of a dataset. In this paper, some efficient measurements for improving the conventional algorithm kernel fuzzy K - means clustering algorithm (KFCM) are proposed. Firstly, we study the parameters initialization, and replace the membership matrix initialization with the centers of clusters initialization which can be obtained by utilizing prior knowledge adequately; secondly, for reducing the time complexity of algorithm, a novel objective function for clustering is proposed based on the centers of classes kernel distance. The simulations demonstrate the feasibility and speedy of the proposed method.