Bayesian Network Structure Learning: A Three-Stage Approach and Its Application

• Activity Number: 190
• Type: Contributed
• Date/Time: Monday, August 10, 2015 : 10:30 AM to 12:20 PM
• Sponsor: Section on Statistical Learning and Data Mining
• Author(s): Kaixian Yu* and Jinfeng Zhang
• Companies: Florida State University and Florida State University
• Keywords: Bayesian networks ; BNs ; Three-stage

Abstract

Bayesian networks (BNs) provide a probabilistic, graphical framework for modeling high-dimensional joint distributions with complicated correlation structures. BNs have wide applications in many disciplines, including biology, social science, finance and biomedical science. Despite extensive studies in the past, network structure learning from data is still an open question in BN research. In this study, we present a three-stage approach incorporating sequential importance sampling optimization (SISO). We propose a strategy which requires less observations to accomplish the skeleton discovery, then utilize SISO to find the optimal networks. Eventually we design a new stage to reclaim as many missed edges. We obtained sounding results from testing on simulation networks and the well-known flow cytometry dataset.