A Multi-Objective Genetic Algorithm with Fuzzy Relational Clustering for Automatic Data Clustering

Abstract

A Fuzzy relational clustering algorithm (FRC) based on multi-objective nondominated Sorting genetic algorithm (NSGA-II) called FRC-NSGA-II is proposed for automatic data clustering. A given data set is spontaneously promoted into an optimal number of groups in a precise fuzzy partition through the fuzzy relational clustering algorithm, FRC. FRC operates on a similarity square matrix which is generated by comparing the pair wise similarities between data points. Multi objective NSGA-II is employed to search for appropriate number of partitions for different cluster shapes. Moreover, two well-known cluster validity indices, compactness and separation, are optimized concurrently through multi-objective NSGA-II where compactness indicates variation between data within a cluster and separation means quantifying the separation between different clusters. Real encoding schema is used for variable length NSGA-II chromosomes representing the variable number of clusters. The simulation result on benchmark data sets exhibits that the proposed method gives promising results in the complex, overlapped, high-dimensional non-gene and gene expression data sets and it has better capability of determining well-separated, hyper spherical and overlapping clusters compared with other existing clustering algorithms.