Locating clusters in noisy data: a genetic fuzzy c-means clustering algorithm

Abstract

The paper investigates the use of a genetic algorithm to locate fuzzy clusters embedded in noisy data. The partitioning of data into clusters is an important problem with many applications. Typically, one locates partitions using an iterative fuzzy c-means algorithm. To overcome some of the shortcomings of fuzzy c-means, a genetic c-means clustering algorithm is implemented and evaluated. It was discovered that this genetic c-means algorithm performs well in the absence of noise. When the clusters are embedded in noise, the genetic algorithm is not as robust as the validity guided robust fuzzy clustering algorithm. The paper concludes with a discussion of what factors contribute to the performance and what modifications may increase the robustness of the genetic c-means algorithm.