Genetic selection of features for clustering and classification

This paper discusses some of the issues involved in feature selection for practical applications. Two problems are introduced: 1) an extension of a standard machine learning problem, and 2) from an industrial application, which is used to investigate the value of the proposed technique. A method is proposed which uses a genetic algorithm to identify groups of features for use in classification or clustering algorithms, using a K-nearest neighbour evaluation function. This has the advantage of being computationally faster than creating new classifiers. The results obtained show that the genetic algorithm is an efficient method of solving the feature selection problem