Fuzzy-rough feature selection and a fuzzy 2-level complementary approach for classification of gene expression data

Abstract

Classification of gene expression data is an important issue in medical diagnosis of disease such as cancer. In this paper first Fuzzy-Rough Set theory is established to select relevant features for classification. This will be followed by proposing a new fuzzy 2-level complementary learning method. The Fuzzy-Rough Set is a mathematical tool which encapsulates the relevant but distinct concepts of fuzziness and indiscernibility. These are caused due to uncertainties in knowledge or datasets. Thus a feature selection using this tool is designed to handle two complementary kinds of uncertainties and to increase the accuracy of the outcome. Complementary learning mechanism, on the other hand, has significant performance because it is responsible for human pattern recognition whose is effective in the learning stage and the problem solving. The proposed classification system works in two levels of different accuracies. If the first level fails to process the sample, the second level would handle. A simulation is carried out using some published datasets. The performance of the proposed classification method by means of achieving an excellent accuracy rate of the classification will be shown significantly with respect to some recently proposed methods. Key words: Gene expression data analysis, fuzzy-rough set feature selection, complementary learning method, hierarchical fuzzy system.