A MultiExpert Approach for Bayesian Network Structural Learning

Abstract

The determination of a Bayesian network structure, especially in the case of wide domains, can be often complex, time consuming and imprecise. Therefore the interest of scientific community in learning Bayesian network structure from data is increasing: many techniques or disciplines, as data mining, text categorization, ontology building, can take advantage from structural learning. In literature there are many structural learning algorithms but none of them provides good results in every case or dataset. This paper introduces a method for structural learning of Bayesian networks based on a Multi-Expert approach. The proposed method combines the outputs of five well known structural learning algorithms according to a majority vote combining rule. This approach shows a performance that is better than any single algorithm. This paper shows an experimental validation of the proposed algorithm on a set of "de facto" standard networks, measuring performance both in terms of the network topological reconstruction and of the correct orientation of the obtained arcs. The first results seem to be promising.