Variable Selection through Genetic Algorithms for Classification Purposes

Abstract

In many real world applications, when coping with classification tasks, the problem of the selection of the variables to be used for the development of any kind of classifier has to be faced. This necessity is normally due to the high number of variables which could be potentially included in the input set combined with the lack of a priori knowledge to support the selection process. In this paper variable selection is achieved by means of the use of GAs through a selection process based on the evaluation of the performance of the possible variable combinations used to train a decision tree. Furthermore the proposed method optimizes some parameters of the employed classifier. Within the proposed method several different approaches have been tested on a real industrial problem. The proposed approaches, which are characterized by different initialization and fitness functions of the GAs, obtain very satisfactory results.