A new robust clustering algorithm-density-weighted fuzzy c-means

Abstract

Presents a robust clustering algorithm called density-weighted fuzzy c-means (DWFCR). Three well-known clustering algorithms, namely, the possibilistic c-means (PCM), the noise clustering (NC), and credibility fuzzy c-means (CFCM) are studied. We observed that the partition performance in these algorithms are sensitive to the changes of memberships. In order to reduce sensitivity to noise and improve the mode-seeking capability, in DWFCM we used a method that incorporates a potential measurement to identify input data before the clustering process. The measurement can faithfully reveal the degree of density around an input data point. Compared to FCM, DWFCM is less sensitive to outliers and noise and has better performance in mode-seeking, while preserving the partition ability of FCM. Performance comparison of DWFCM and these algorithms are given.