Efficient Markov Blanket Discovery and Its Application.

Abstract

In a Bayesian network (BN), a target node is independent of all other nodes given its Markov blanket (MB), and finding the MB has many applications, including feature selection and BN structure learning. We propose a new MB discovery algorithm, simultaneous MB (STMB), to improve the efficiency of the existing topology-based MB discovery algorithms. The proposed method removes the necessity of enforcing the symmetry constraint that is prevalent in existing algorithms, by exploiting the coexisting property between spouses and descendants of the target node. Since STMB mainly reduces the number of independence tests needed to complete the MB set after finding the parents-and-children set, it is applicable to all previous topology-based methods. STMB is both sound and complete. Experiments show that STMB has a comparable accuracy but much better efficiency than state-of-the-art methods. An application on benchmark feature selection datasets further demonstrates the excellent performance of STMB.