Feature Selection for Efficient Local-to-Global Bayesian Network Structure Learning

Local-to-global learning approach plays an essential role in Bayesian network (BN) structure learning. Existing local-to-global algorithms first construct the skeleton of a DAG (directed acyclic graph) by MB (Markov blanket) or PC (parents and children) each variable data set, then orient edges skeleton. However, existing methods are often computationally expensive especially with large-sized BN, resulting inefficient algorithms. To tackle problem, this paper, we link feature selection local BN develop efficient using filtering selection. Specifically, analyze rationale well-known Minimum-Redundancy Maximum-Relevance (MRMR) for set variable. Based on analysis, propose F2SL (feature selection-based learning) to The employs MRMR learn DAG, orients Employing independence tests score functions orienting edges, instantiate into two new algorithms, F2SL-c (using tests) F2SL-s functions). Compared state-of-the-art experiments validated that proposed paper more provide competitive quality than compared