Biased Random-Key Genetic Algorithm for Structure Learning

Abstract

The structure learning of Bayesian networks is a NP-hard problem, which cannot be easily solved since it is usually a complex combination optimization problem. Thus, many structure learning algorithms using evolutionary techniques are investigated recently to obtain a reasonable result. However, evolutionary algorithms may suffer from a low accuracy and restricts their applications. In this paper, we apply the Biased Random-Key Genetic Algorithm to solve Bayesian network structure learning problem since this framework is novely designed to solve conventional combination optimization problems. Also, we use a local optimization algorithm as its decoder to improve the performance. Experiments show that our method achieves better performances on the real-world networks than other state-of-art algorithms.