Bayesian network structure learning based on improved particle swarm optimization

Bayesian network structure learning is one of the important research techniques in the domain of data mining and knowledge discovery, when the search space of the network structure is bigger, traditional binary particle algorithms often have some defects such as low convergent speed, falling easily into local optimum and low precision. We improve the classic binary particle swarm optimization algorithm in two respects: particle initialization and update process; the improved algorithm has stronger optimization ability. We compare the proposed algorithm with the original algorithm using the ASIA network. The results and their analysis show preliminarily that the proposed algorithm is able to find the better solution with less number of iterations, without increasing the complexity basically.