Improving Interpretability of Fuzzy Models Using Multi-Objective Neuro-Evolutionary Algorithms
Gracia Snchez, Jos Francisco Snchez Ruiz, Jos Manuel Alcaraz Muoz, Fernando Jimnez · InTech eBooks · 2008
This chapter remarks on some results in the combination of Pareto-based multi-objective evolutionary algorithms, neural networks and fuzzy modeling. A multi-objective constrained optimization model is proposed in which criteria such as accuracy, transparency and compactness have been taken into account. Three multi-objective evolutionary algorithms (MONEA, ENORA-II and NSGA-II) have been implemented in combination with neural network based and rule simplification techniques. The results obtained improve on other more complex techniques reported in literature, with the advantage that the proposed technique identifies a set of alternative solutions. Statistical tests have been performed over the hypervolume quality indicator to compare the algorithms and it has shown that, for the non linear plant problem, ENORA-II obtains better results than MONEA and NSGA-II algorithms. Future improvements of the algorithms will be the automatic parameter tuning, and a next application of these techniques will be on medicine data.