Kernel Subspace Possibilistic Fuzzy C-Means Algorithm Driven by Feature Weights

Abstract

AbstractAt present, one of the difficulties in the field of fuzzy clustering is the clustering analysis for high dimensional data. Most of the existing fuzzy clustering algorithms are sensitive to initialization, which are greatly affected by noise points and has weak adaptability to high-dimensional data. In order to solve these problems, a kernel space possibilistic fuzzy c-means clustering (KSPFCM for short) algorithm is proposed. Firstly, considering that the fuzzy possibilistic clustering algorithm can deal with the noise points better, the typicality matrix is introduced to weaken the constraint on the membership matrix. Secondly, to refine the granularity of clustering convergence, subspace clustering is introduced to assign each feature weight value to the data, and the feature weight value is obtained by adaptive collaborative iteration. Among them, a greater weight value are given for the important features of the data, which makes the feature weight allocation more reasonable. Thirdly, the Gaussian kernel distance is used as the distance measure between data points to optimize the sensitivity to the data set, and thus the KSPFCM algorithm is proposed. Finally, by comparing with several advanced clustering algorithms on the artificial data sets and UCI data sets in this field, the proposed KSPFCM algorithm performs better in the five clustering effectiveness indicators of ACC, EARI, NMI, CHI and XBI.KeywordsCollaborative intelligenceFuzzy clusteringSubspacePossibilistic clusteringKernel functions