Fuzzy clustering with fitness predator optimizer for multivariate data problems

Abstract

Fuzzy c-means (FCM) is the most common fuzzy clustering model and uses an objective function to measure the desirability of partitions. However, if the data sets contain several noise points, or if the data sets are very high dimensional, the iteration process of optimization the FCM model often falls into local optima solution. To avoid this problem, this paper proposes a new hybrid fuzzy clustering algorithm that incorporates the Fitness Predator Optimizer (FPO) into the FCM model. FPO is a new bionic-inspired algorithm to avoid premature convergence for the multimodal optimization problem. The excellent probability of finding the global optimum of FPO enhances the quality of fuzzy clustering. Five benchmark data sets from the UCI Machine Learning Repository are used to compare the performances of proposed FPO-FCMwith FCMand a hybrid swarm algorithm based on Quantum-behaved PSO. Experimental results show that the proposed approach could demonstrate the desirable performance and avoid the minimum local value of objective function for multivariate data type clustering problems.