Bi-Objective Search Method for Bayesian Network Structure Learning

Abstract

Bayesian network (BN) is a probability graph model, which makes uncertain reasoning logically clearer and more understandable. Structure learning is the first step to learn a BN model. And the score + search methods are a kind of the effective methods to learn the structure. This paper proposes a Bi-Objective Search (BOS) method for Bayesian network structure learning, which considers two objectives, i.e., the log-likelihood score and network complexity. To avoid the illegal structures, BOS samples edges and generates permutations to add directions to the edges for the initial population. To improve the diversity, BOS designs the genetic operators to generate new solutions. The new approach is applied to a set of discrete Bayesian networks, and the experimental results show that the algorithm is superior to the existing algorithms in BN structure learning.