Hybrid Fuzzy Based Nature-inspired Clustering Algorithms with Validity Measures

Abstract

Data Clustering means the act of partitioning a data set into group of similar objects. It is an important activity of data mining. Fuzzy c-means (FCM) is one of the most widely used data clustering methods. However, it suffers from some limitations like easily struck in local minima and sensitive to noise and outlier. Fuzzy possibilistic c-means (FPCM) algorithm is one of the good methods for noisy environment. Partition Index Maximization (PIM) is one of the extensions of FCM algorithm by adding partition coefficient (PC) into FCM objective function. Nature-inspired algorithm such as particle swarm optimization (PSO) is a global optimization technique. It overcomes the problem of local optima. The performance of PSO algorithm can be further improved with the help of fuzzy clustering algorithms. Most of the traditional clustering algorithms are based on Euclidean distance measure. In this paper, two hybrid algorithms namely fuzzy possibilistic c-means based particle swarm optimization algorithm (FPCM-PSO) and fuzzy c-means based particle swarm optimization algorithm with PIM (FUZZY-PSO-PIM) are proposed using different measures including Euclidean, Manhattan and Chessboard distance. Experimental results on well-known real world benchmark UCI repository biomedical data sets and an artificial data set show that Fuzzy-PSO-PIM hybrid algorithm is efficient and report encouraging results than other clustering techniques for all the distance measures. The clustering results are evaluated with respect to many cluster validity measures. It is also observed that hybrid algorithms based on chessboard distance measure produce better results than the other distance measures.

Copyright © 2014 Praise Worthy Prize - All rights reserved.