Analysis of the impact of using different diversity functions for the subgroup discovery algorithm NMEEF-SD

Abstract

A main purpose of a multi-objective evolutionary algorithm is to find a good relationship between convergence and diversity of the population. Convergence guides the algorithm to search the optimal solution and diversity tries to avoid a premature stagnation of the search. In multi-objective evolutionary algorithms, diversity has been promoted using different techniques. In this paper, several diversity functions were implemented in NMEEF-SD, an algorithm for the extraction of fuzzy rules in a subgroup discovery task, to analyse the influence of these functions in the evolutionary process. The results show the advantages of the different measures, depending on the intended objective.