Aprendizagem estrutural de redes bayesianas utilizando algoritmo genético multi-agente

Abstract

This work approaches the structural learning of Bayesian networks. The task of learning a Bayesian Network from data is an Np-hard problem. We present the Multi-Agent Genetic Algorithm (MAGA) to learn Bayesian network structures. The MAGA algorithm uses the elements of a multi-agent system, such as communication between agents, their interaction with the environment, and the mechanisms of genetic algorithms that subject agents to genetic operators. In this study, two configurations of MAGA were implemented, which were compared to other algorithms in the literature. The purpose of this research is to evaluate the performance of the MAGA, mainly in instances with many variables and large volumes of data. The experiment results show that the two configurations of MAGA are effective in learning the Bayesian Network structures, chiefly in instances with many variables and large volumes of data.