Employing effective feature selection in Genetic Fuzzy Rule-Based Classification Systems

This paper proposes the use of a local feature selection scheme, for the effective selection of relevant features, when designing Genetic Fuzzy Rule-Based Classification Systems (GFRBCSs). The method relies in providing the genetic search with deterministic information about the quality of each feature with respect to its classification ability, directing the evolution in selecting the most useful features. To evaluate our method, we propose a learning algorithm that iteratively generates the final fuzzy rule base, extracting one rule at a time, as directed by a boosting algorithm. Experimental results in a number of well-known classification datasets prove the efficiency of the proposed system in dealing with high-dimensional feature spaces.