FRvarPSO: A method for obtaining fuzzy classification rules using optimization Techniques

Abstract

FRvarPSO is a new method for obtaining classification rules, which operates on nominal and/or fuzzy attributes. It combines LVQ, which is a supervised learning neural network. The search is performed through an optimization technique such as varPSO, considered a metaheuristic based on particles clusters of variable population. Each individual represents a possible solution to the problem. The proposed method uses a voting criterion, which affects the particle’s speed. This method is benchmarked against PART and C4.5, on 12 databases of the UCI repository and three databases of financial institutions from Ecuador. The results obtained were satisfactory.