Rough-fuzzy granular computing with application to bioinformatics and web intelligence

Components of machine intelligence and and relation with soft computing are explained. The role of rough fuzzy computing in the said framework along with the relevance of integration is explained. Two examples of such integration are described for problems of clustering and generating class prototypes. Here rough sets are used for granular computing using information granules and for uncertainty handling in defining cluster shapes using lower and upper approximation, whereas fuzzy sets are used for linguistic representation and fuzzy granulation, and modeling uncertainty arising from overlapping regions. In effect, rough-fuzzy clustering provides a balanced compromise between restrictive (hard) and descriptive (fuzzy) representations of class belonging for overlapping regions. The use of rough sets as an ensemble classifier is then described. The ensemble of the base classifiers is optimum in the sense that the performance of the resulting classifier is at least equal to that of the best base classifier. The ensemble classifier is shown to be effective in classifying web documents and services. The significance of rough-fuzzy case generation is demonstrated in terms of case generation, case retrieval, average no. of features required per case and classification performance. The features and merits of rough-fuzzy clustering are demonstrated for determining the bio-bases in encoding protein sequence for analysis where the biobases coorrespond to c-medoids. New quantitative indices are stated in this context. The talk concludes with their future challenges and possible uses in data mining.