A novel binary adaptive differential evolution algorithm for Bayesian Network learning

Bayesian Network is the most popular method for uncertain expert knowledge and ratiocination, and wildly applied in large number of research area. The primary strategy for Bayesian Network learning is to select the optimal network candidates by using statistical score. In this paper, we propose a novel Binary Differential Evolution algorithm for Bayesian Network learning (BINDEBN). BINDEBN adopts an adaptive 0/1 matrix as the scale factor, and implements the information exchange among Bayesian Networks during learning process by crossover and mutation operators. Then, BINDEBN selects the Bayesian Network candidates from network model space according to Bayesian Information Criterion (BIC) scoring. The experiment results prove that the excellent performance of our method.