An improved multi-objective evolutionary optimization of data-mining-based fuzzy decision support systems

The paper presents an approach to designing from data fuzzy decision systems (fuzzy rule-based classifiers (FRBCs)) by means of four multi-objective evolutionary optimization algorithms (MOEOAs) including the well-known NSGA-II, epsilon-NSGA-II, SPEA2, and our generalization of SPEA2 (referred to as SPEA3). The advantages of SPEA3 (a better-balanced distribution and a higher spread of solutions than for SPEA2) are shown using selected benchmark tests. The main building blocks of our FRBC and the main components of its MOEOA-based optimization are briefly presented. The proposed FRBCs with genetically optimized accuracy-interpretability trade-off are effective and modern tools for intelligent decision support in various areas of applications. In this paper, the application to designing credit-granting decision support system based on Statlog (German Credit Approval) financial benchmark data set is presented. A comparison of our approach employing various MOEOAs is also carried out.