Bayesian network structure learning based on the chaotic particle swarm optimization algorithm

The Bayesian network (BN) is a knowledge representa­tion form, which has been proven to be valuable in the gene regulatory network reconstruction because of its capability of capturing causal re­lationships between genes. Learning BN structures from a database is a nondeterministic polynomial time (NP)-hard problem that remains one of the most exciting challenges in machine learning. Several heuristic searching techniques have been used to find better network structures. Among these algorithms, the classical K2 algorithm is the most suc­cessful. Nonetheless, the performance of the K2 algorithm is greatly af­fected by a prior ordering of input nodes. The proposed method in this paper is based on the chaotic particle swarm optimization (CPSO) and the K2 algorithm. Because the PSO algorithm completely entraps the lo­cal minimum in later evolutions, we combined the PSO algorithm with the chaos theory, which has the properties of ergodicity, randomness, and regularity. Experimental results show that the proposed method can improve the convergence rate of particles and identify networks more efficiently and accurately.