AIFSA: A New Approach for Feature Selection and Weighting

Abstract

Feature selection is a typical search problem where each state in the search space represents a subset of features candidate for selection. Out of n features, 2n subsets can be constructed, hence, an exhaustive search of all subsets becomes infeasible when n is relatively large. Therefore, Feature selection is done by employing a heuristic search algorithm that tries to reach the optimal feature subset. Here, we propose a new wrapper feature selection and weighting algorithm called Artificial Immune Feature Selection Algorithm (AIFSA); the algorithm is based on the metaphors of the Clonal Selection Algorithm (CSA). AIFSA, by itself, is not a classification algorithm, rather it utilizes well-known classifiers to evaluate and promote candidate feature subset. Experiments were performed on textual datasets like WebKB and Syskill&Webert web page ratings. Experimental results showed AIFSA competitive performance over traditional well-known filter feature selection approaches as well as some wrapper approaches existing in literature.