Towards Big Data Bayesian Network Learning - An Ensemble Learning Based Approach
Yan Tang, Yu Wang, Kendra Cooper, Ling Li · 2014
Recently, we are entering the Big Data era[[1]]. The Bayesian Network (BN), as a directed probabilistic graph model, is providing intuitive knowledge presentation and accurate prediction for many mission critical areas. However, the current algorithms do not scale well for Big Data Bayesian network learning. This paper proposes a novel parallel BN learning algorithm called PENBays (Parallel ENsemble based Bayesian Networks Learning), which integrates the best BN learning algorithms MMHC, TPDA and REC. It has three phases: Data Preprocess (DP), Individual Ensemble Learning (IEL) and Central Ensemble Learning (CNL). Through these phases, PENBays effectively learns a BN rapidly from large datasets. Experiments reveal that PENBays learns BNs with better accuracy than base line learning algorithms like MMHC, TPDA and REC, showing promising application potential in the big data mining area.