Parameters learning of BN in small sample base on data missing

Introduced the support vector machines regression and put forward the Bayesian networks parameters learning algorithm with data repaired function. The algorithm makes use of the observed information of each observation node on the Bayesian networks in different times, without any constraints of priori information, to repair the missing data by the sample regression. On the basis of the completed data obtained, the algorithm uses the maximum likelihood estimation to estimate the Bayesian network parameters. The simulation result indicates that under the situation of missing small sample data, compared with the standard EM algorithm, the parameter learning method can increase the efficiency of the parameter learning and improve the precision of the inference.