Applying Restrained Genetic Algorithm for Attribute Reduction Using Attribute Dependency and Discernibility Matrix

Abstract

Microarray gene dataset often contains huge number of attributes many of which are irrelevant and redundant with respect to classification. Presence of such attributes may sometimes reduce the classification accuracy of the dataset. Therefore, the data should be pre-processed to filter out the unimportant attributes before passing them on to the classifier. In the paper, the concepts of Rough Set Theory (RST) and Genetic Algorithm (GA) are used for selecting only the relevant attributes of the dataset. The method constructs relative discernibility matrix to compute the core attributes based on which attributes are encoded to strings used as an initial population for running the genetic algorithm. The method runs each time by adding a single attribute to the initial strings to select only a minimal attribute set known as reduct. The fitness function is defined based on the attribute dependency of the formed rough set. Attribute dependency gives a measure of the degree of influence of the selected attribute subset on the decision. The experimental results show that, the proposed method yields better result than some well-known attribute reduction algorithms for some real-world microarray cancerous datasets.