Learning fuzzy classification rules from imbalanced datasets using multi-objective evolutionary algorithm

Fuzzy systems have been used to solve different types of problems, for example, classification problems. Genetic algorithms are a type of evolutionary algorithms used to automatically learn or tune components of the fuzzy systems from data. Recently multi-objective evolutionary algorithms have been used for this task, since they can consider multiple conflicting objectives, for example, accuracy and interpretability which are desirable properties of the fuzzy systems. Learning rules from imbalanced datasets is considered a research trend in this area. This work proposes a method to learn fuzzy classification rules from imbalanced datasets using multi-objective genetic algorithms and the iterative rule learning approach. In this approach, a single rule is learnt in each execution of the multi-objective evolutionary algorithm. The proposed method contains two phases: (i) pre-processing, to balance the imbalanced dataset; (ii) iterative fuzzy rule learning using multi-objective evolutionary algorithms. The algorithm uses two objectives: the accuracy and number of conditions of each fuzzy rule; the method proposed here is an extension of a method previously proposed by the authors. The results show that the new proposed method has a good performance. The obtained accuracy and the number of conditions are better than other genetic method used in the comparison analysis.