A fast and efficient multi-objective evolutionary learning scheme for fuzzy rule-based classifiers

摘要: During the last years, multi-objective evolutionary algorithms (MOEAs) have been extensively used to generate fuzzy rule-based systems characterized by different trade-offs between accuracy and complexity. In this paper, we propose an MOEA-based approach learn concurrently rule data bases of classifiers (FRBCs). particular, are generated exploiting a condition selection (RCS) strategy, which selects reduced number rules from heuristically set candidate conditions for each selected during process. RCS can be considered as learning in constrained search space. As regards base learning, membership function parameters linguistic term learned application RCS. We tested our on twenty-four classification benchmarks compared results with ones obtained two similar state-of-the-art approaches well-known non-evolutionary algorithms, namely FURIA C4.5. Using non-parametric statistical tests, show that generates FRBCs complexity statistically comparable to, sometimes better than, approaches, exploiting, however, only 5% fitness evaluations these approaches. Further, result more interpretable than C4.5 while achieving same level.