An Efficient Kernelized Fuzzy Possibilistic C-Means for High-Dimensional Data Clustering

Abstract

Clustering high-dimensional data has been a major concern owing to the intrinsic sparsity of the data points. Several recent research results signify that in case of high-dimensional data, even the notion of proximity or clustering possibly will not be significant. Fuzzy c-means (FCM) and possibilistic c-means (PCM) have the capability to handle the high-dimensional data, whereas FCM is sensitive to noise and PCM requires appropriate initialization to converge to nearly global minimum. Hence, to overcome this issue, a fuzzy possibilistic c-means (FPCM) with symmetry-based distance measure has been proposed which can find out the number of clusters that exist in a dataset. Also, an efficient kernelized fuzzy possibilistic c-means (KFPCM) algorithm has been proposed for effective clustering results. The proposed KFPCM uses a distance measure which is based on the kernel-induced distance measure. FPCM combines the advantages of both FCM and PCM; moreover, the kernel-induced distance measure helps in obtaining better clustering results in case of high-dimensional data. The proposed KFPCM is evaluated using datasets such as Iris, Wine, Lymphography, Lung Cancer, and Diabetes in terms of clustering accuracy, number of iterations, and execution time. The results prove the effectiveness of the proposed KFPCM.