Improved bayesian optimization algorithm incorporated local structure learning

Learning the Bayesian networks is a key for a successful application of Bayesian optimization algorithm. However it is NP-hard to learn the Bayesian networks. In order to get the reliable Bayesian networks quickly, a novel learning strategy is presented in this paper. Firstly, the stochastic greedy algorithm for local structure is introduced according to the decomposable of scoring metric. The optimal edge is selected by using scoring metric and local search. Secondly, by analysis it is conclude that the reliability of the Bayesian networks which is learned by the proposed algorithm is improved. Due to the reliable network, BOA overcomes deceptive and performs efficiently. Experimental results show the improved algorithm's performances are better than those of traditional BOA.