基于马尔科夫毯约束的最优贝叶斯网络结构学习算法

To solve the problem about structure learning of optimal Bayesian network, this paper proposes Dynamic Programming Constrained with Markov Blanket (DPCMB), which uses Markov Blanket calculated by Incremental Association Markov Blanket (IAMB) to reduce the number of scoring calculations in Dynamic Programming. We research on the effect of the significance value in IAMB on the performance indicators of DPCMB algorithm, and give reasonable suggestions for adjusting the significance value. Experimental results show that the DPCMB algorithm can adjust the significance value so that the accuracy of the algorithm is comparable to that of the DP algorithm, and running time, score calculation times, and memory requirements of the algorithm are greatly reduced.