Genetic algorithm with fuzzy fitness function for feature selection

Feature selection is an important preprocessing task for any pattern recognition or data mining application. Though lots of well developed statistical and mathematical techniques of feature selection exist they do not match the imprecise and incomplete nature of most of the real world problems. Recently soft computing techniques i.e. neurocomputing, fuzzy logic, genetic algorithm etc. are gaining growing popularity for their remarkable ability of handling real life data like a human being in an environment of uncertainty, imprecision and implicit knowledge. In this work, a genetic algorithm in conjunction with a fuzzy fitness function, a fuzzy measure for evaluation of the quality of a feature has been proposed for feature subset selection. GA based feature selection algorithms are robust but their computation time is high specially when they are used with a classifier for fitness evaluation. The computationally light fuzzy fitness function lowers the computation time of the traditional GA based algorithm with classifier accuracy as the fitness function by separating the two stages feature selection and classification. Simulation over two data sets shows the efficiency of the proposed technique for achieving near optimal solution in practical problems specially when the data set contains a large number of features.