A Novel Framework to Design Fuzzy Rule-Based Ensembles Using Diversity Induction and Evolutionary Algorithms-Based Classifier Selection and Fusion

Fuzzy rule-based systems have shown a high capability of knowledge extraction and representation when modeling complex, nonlinear classification problems. However, they suffer from the so-called curse of dimensionality when applied to high dimensional datasets, which consist of a large number of variables and/or examples. Multiclassification systems have shown to be a good approach to deal with this kind of problems. In this contribution, we propose an multiclassification systembased global framework allowing fuzzy rule-based systems to deal with high dimensional datasets avoiding the curse of dimensionality. Having this goal in mind, the proposed framework will incorporate several multiclassification system methodologies as well as evolutionary algorithms to design fuzzy rule-based multiclassification systems. The proposed framework follows a two-stage structure: 1) fuzzy rule-based multiclassification system design from classical and advanced multiclassification system design approaches, and 2) novel designs of evolutionary component classifier combination. By using our methodology, different fuzzy rule-based multiclassification systems can be designed dealing with several aspects such as improvement of the performance in terms of accuracy, and obtaining a good accuracy-complexity trade-off.