Markov blanket-embedded PSO algorithm for Bayesian network structure learning

In the Bayesian network (BN) structure search process, due to the uncertainty of the initial population, the learning efficiency in the structural optimization process is reduced, and the convergence speed is slowed down. We propose a Markov blanket-embedded particle swarm optimization (PSO) algorithm for BN structure learning in our research. In particular, the efficient and effective discovery of the Markov blanket method is used to obtain the undirected network framework. A local topology is established for each node to implement the initialization. Parameter adaptive adjustment is carried out through an improved PSO algorithm. Initial population distribution information is adjusted through Algorithm parameters. For improved structure, a search efficiency mutation mechanism is added for increased diversity. The experimental results show that our algorithm can get better search accuracy and faster convergence speed than other algorithms.