Rule extraction from a neural network by hierarchical multiobjective genetic algorithm

Abstract

The paper presents a method of rule extraction from the trained neural network by means of a genetic algorithm. The multiobjective approach is used to suit the nature of the problem, since different criteria (accuracy, complexity) may be taken into account during the search for a satisfying solution. The use of a hierarchical algorithm aims at reducing the complexity of the problem and thus enhancing the method performance. The overall structure and details of the algorithm as well as the results of experiments performed on popular benchmark data sets are presented.