DBN structure learning based on MI-BPSO algorithm

摘要

To improve the accuracy of structure learning for Dynamic Bayesian Network (DBN), this paper proposes Mutual Information-Binary Particle Swarm Optimization (MI-BPSO) algorithm. The MI-BPSO algorithm firstly uses MI and conditional independence test to prune the search space and speed up the convergence of the searching phase, then calls BPSO algorithm to search the constrained space and get the intra-network and inter-network of DBN. Experimental results show that this algorithm performs as well as K2 while it doesn't need a given variable ordering, and performs better than MWST-GES, MWST-HC and I-BN-PSO.