Título: Score Metrics for Learning Bayesian Networks used as Fitness Function in a Genetic Algorithm
Autores: Santos, Edimilson B. dos; Hruschka Jr., Estevam R.; Ebecken, Nelson F. F.
Resumo: Variable Ordering (VO) information may be used as a constraint to reduce the search space of learning Bayesian Networks (BNs) from data. Several authors have proposed the use of Evolutionary Algorithms (EAs) to find such an ordering. In this work, a genetic algorithm named VOGA (Variable Ordering Genetic Algorithm) has been applied to this goal. Since the fitness function plays an important role in the performance of genetic algorithms, we present and discuss five score metrics, applied to induce BNs from data, to be used as fitness function in VOGA. The main objective is to investigate the VOGA performance when employing different metrics.
Código DOI: 10.21528/CBIC2011-38.2
Artigo em pdf: st_38.2.pdf
Arquivo BibTex: st_38.2.bib