A Hybrid Optimization Algorithm for Bayesian Network Structure Learning Based on Database

The process of learning Bayesian networks includes structure learning and parameters learning. During the process, learning the structure of Bayesian networks based on a large database is a NP hard problem. The paper presents a new hybrid algorithm by integrating the algorithms of MMPC (max-min parents and children), PSO (particle swarm optimization) and GA (genetic algorithm) effectively. In the new algorithm, the framework of the undirected network is firstly constructed by MMPC, and then PSO and GA are applied in score-search. With the strong global optimization of PSO and the favorable parallel computing capability of GA, the search space is repaired efficiently and the direction of edges in the network is determined. The proposed algorithm is compared with conventional PSO and GA algorithms. Experimental results show that the proposed algorithm is most effective in terms of convergence speed.