Learning Using Hybrid Intelligence Techniques

Abstract

This chapter focuses on a few key applications of hybrid intelligence techniques in the field of feature selection and classification. Hybrid intelligent techniques have been used to develop an effective and generalized learning model for solving these applications. The first application employs a new evolutionary hybrid feature selection technique for microarray datasets, which is implemented in two stages by integrating correlation-based binary particle swarm optimization (BPSO) with rough set algorithm to identify non-redundant genes capable of discerning between all objects. The other applications discussed are for evaluating the relative performance of different supervised classification procedures using hybrid feature reduction techniques. Correlation based Partial Least square hybrid feature selection method is used for feature extraction and the experimental results show that Partial Least Squares (PLS) regression method is an appropriate feature selection method and a combined use of different classification and feature selection approaches make it possible to construct high performance classification models for microarray data. Another hybrid algorithm, Correlation based reduct algorithm (CFS-RST) is used as a filter to eliminate redundant attributes and minimal reduct set is produced by rough sets. This method improves the efficiency and decreases the complexity of the classical algorithm. Extensive experiments are conducted on two public multi-class gene expression datasets and the experimental results show that hybrid intelligent methods are highly effective for selecting discriminative genes for improving the classification accuracy. The experimental results of all the applications indicate that, all the hybrid intelligent techniques discussed here have shown significant improvements in most of the binary and multi-class microarray datasets.