Hybrid Bayesian network discovery with latent variables by scoring multiple interventions

Abstract In Bayesian Networks (BNs), the direction of edges is crucial for causal reasoning and inference. However, Markov equivalence class considerations mean it not always possible to establish edge orientations, which why many BN structure learning algorithms cannot orientate all from purely observational data. Moreover, latent confounders can lead false positive edges. Relatively few methods have been proposed address these issues. this work, we present hybrid mFGS-BS (majority rule Fast Greedy Search with Scoring) algorithm discrete data that involves an set one or more interventional sets. The assumes insufficiency in presence variables produces a Partial Ancestral Graph (PAG). Structure relies on approach novel scoring paradigm calculates posterior probability each directed being added learnt graph. Experimental results based well-known networks up 109 10 k sample size show improves accuracy relative state-of-the-art computationally efficient.