A hybrid kernel-based possibilistic fuzzy c-means clustering and cuckoo search algorithm

Abstract

Possibilistic Fuzzy c-means (PFCM) algorithm is a robustness clustering algorithm that combines two algorithms, Fuzzy c-means (FCM) and Possibilistic c-means (PCM). It addresses the weakness of FCM in handling noise sensitivity and the weakness of PCM within the case of coincidence clusters. However, PFCM works inefficiently when the input data is nonlinear separable. To solve this problem, kernel methods have been introduced into possibilistic fuzzy c-means clustering (KPFCM). KPFCM can address noises or outliers data better than PFCM. But KPFCM suffers from a common drawback of clustering algorithms that may be trapped in local minimum which results in not good results. Recently, Cuckoo search (CS) based clustering has proved to achieve fascinating results. It can achieve the best global solution compared to most other metaheuristics. In this paper, we propose a hybrid method encompassing KPFCM and Cuckoo search algorithm to form the proposed KPFCM-CSA. The experimental results indicate that the proposed method outperformed various well-known recent clustering algorithms in terms of clustering quality.