Efficient score-based Markov Blanket discovery

doi:10.1016/j.ijar.2016.09.009

2017-01-01

Tian Gao *

Qiang Ji

DOI：10.1016/j.ijar.2016.09.009

原文链接

查看原文

摘要

En 中文

In a Bayesian Network (BN), the Markov Blanket (MB) of a target node consists of its parents, children, and spouses, and the target node is independent of all other nodes given its MB. Finding the MB has many applications, including feature selection and BN structure learning. We propose two new Markov Blanket discovery algorithms, score-based Simultaneous Markov Blanket discovery ((STMB)-T-2) and its more efficient variant, (STMB)-T-2+, to improve the efficiency of existing score-based MB learning algorithms. The proposed methods remove the necessity of enforcing the commonly used symmetry constraint by exploiting the coexistence property between spouses and descendants of the target node. (STMB)-T-2 and (STMB)-T-2+ achieve comparable accuracy and better efficiency than state-of-theart score-based methods. (STMB)-T-2 and (STMB)-T-2+ are proven sound and complete under one conjecture. Empirical results on standard MB discovery datasets demonstrate the superior performances of (STMB)-T-2 and (STMB)-T-2+. (C) 2016 Elsevier Inc. All rights reserved.

Keyword:

Markov Blanket discovery

Bayesian network

Local structure learning