DefinePK

Abstract

Feature selection helps to improve prediction quality, reduce the computation time, complexity of the model and build models that are easily understandable. Feature selection removes the irrelevant and redundant features and selects the relevant and useful features that provide an enhanced classification results as the original data. This research work analysis the performance of the clustering and genetic algorithm based feature selection (CLUST-GA-FS) algorithm. The proposed algorithm CLUST-GA-FS has three stages namely irrelevant feature removal, redundant feature removal, and optimal feature generation. The algorithm involves removing the irrelevant features, removing redundant features by constructing a minimum spanning tree, splitting the minimum spanning tree into cluster, finding the representative feature from each cluster and finally finding the optimal set of features using genetic algorithm. CLUST-GA-FS algorithm is compared with the existing filter feature selection methods Fast correlation based feature selection (FCBF), Correlation based feature selection (CFS), Information gain (Infogain) and ReliefF. The work uses three microarray dataset Leukemia, Colon and Arcene that are high dimensional.