Learning Bayesian network structures by estimation of distribution algorithms : An experimental analysis

Gregory Thibault (1) , Stéphane Bonnevay (2) , Alexandre Aussem (1)

1 LIESP - Laboratoire d'Informatique pour l'Entreprise et les Systèmes de Production
2 LIRIS - Laboratoire d'InfoRmatique en Image et Systèmes d'information

Learning the structure of a Bayesian network (BN) from a data set is NP-hard. In this paper, we discuss a novel heuristic based on Estimation of Distribution Algorithms (EDA), a new paradigm for Evolutionary Computation that is used as a search engine in the BN structure learning problem. The purpose of this work is to study the parameter setting of the EDA and to fix a "good" set of parameters. For this purpose, the EDA-based procedure is applied on several benchmarks to recover the original structure from data. The quality of the learned structure is assessed using several performance indexes.

Domaines: Réseau de neurones [cs.NE]; Autres [stat.ML]; Intelligence artificielle [cs.AI]

Gregory Thibault, Stéphane Bonnevay, Alexandre Aussem. Learning Bayesian network structures by estimation of distribution algorithms : An experimental analysis. IEEE International Conference on Digital Information Management (ICDIM 07), 2007, Lyon, France. pp.127-132. ⟨hal-00264027⟩