Evolutionary multi-objective granular computing classifiers

Abstract

The classification error rate and the number of granules are two important objectives in granular computing. As two conflict objectives, optimizing them simultaneously is impossible. Evolutionary multi-objective granular computing classifiers are proposed to seek the tradeoff between the minimal classification error rate and the minimal number of granules. The individual is represented as the two-layer structure, the first layer is composed of the sequence of granule, and the second layer includes the beginning points, the end point, and the class labels of granules. Importance-based Pareto (IPareto) dominance is used to the comparison of two individuals. Crossover operation, union operation, and mutation operation designed specially for Granular Computing are performed the evolution process. Compared with Pareto front, IPareto front corresponded to more classifiers for two-class problems and multi-class problems.