Optimizing the induction of Bayesian Networks using PC and Variable Ordering Genetic Algorithms

Variable Ordering (VO) plays an important role when inducing Bayesian Networks (BN). Previous works in the literature suggest that it is worth pursuing the use of genetic algorithms for identifying a suitable VO, when learning a BN structure from data. However, these algorithms may be computationally costly. This paper proposes a hybrid adaptive algorithm named PC-VOGA where initially the PC algorithm is performed to provide a previous VO before the genetic algorithm begins. Such previous ordering is produced by CGSort algorithm, which is also proposed in this work from BN structure induced by PC. Initial experiments revealed that the PC-VOGA approach is promising having the assistance of CGSort algorithm.