Incremental learning Bayesian network structures efficiently
Da Shi, Shaohua Tan · 2010
In this paper, a new hybrid incremental learning algorithm for Bayesian network structures is proposed. It develops a polynomial-time constraint-based technique to build up a candidate parents set for each domain variable, and a hill climbing search procedure is then employed to refine the current network structure under the guidance of those candidate parents sets. Our algorithm always offers considerable computational complexity savings while obtaining better model accuracy compared to existing incremental algorithms when dealing with complex real-world problems. The more complex the real-world problems are, the more significant the advantage our algorithm keeps is.