A soft computing model for data clustering and application to gene grouping

Abstract

Data clustering aims at discovering groups and identifying patterns in data. A large number of clustering algorithms and their variations exist in literature. In this work, we consider data that has a natural ordering based on some criterion. The problem can be stated as clustering of sequential data based on multiple features. It belongs to the class of grouping problems. When pre-ordered data is clustered, it results in contiguous blocks. In the general clustering problem, an all-against-all comparison of data objects is required. However, in sequential data clustering, the data objects are position dependent which imposes the condition that only data objects appearing close together in the data stream will belong to the same cluster, thus avoiding an all-against-all comparison. The classical approach to data clustering has given rise to a large number of algorithms that mainly fall into the hierarchical and partitional categories. Soft Computing paradigms of genetic algorithms, fuzzy logic and artificial neural networks have also contributed towards data clustering. Hybrid algorithms mix different computing families to evolve algorithms that perform better than their constitutive elements. The proposed soft computing model belongs to the class of hybrid algorithms. It draws upon the capabilities of genetic algorithms and fuzzy logic to design a methodology to partition the data set into clusters based on the contribution of a set of factors that are known to have some influence in the formation of clusters. The model consists of a fuzzy guided genetic algorithm based on multiple criteria/features. The model has been successfully applied to the problem of gene grouping in the area of bioinformatics. An organism s genome consists of a sequence of genes. The algorithm attempts to discover groups of related genes that lie adjacent on the genome.