Learning Bayesian network parameters under dual constraints from small data set

In this paper, a novel dual constraints based parameter learning algorithm is presented to overcome the problem of Bayesian network (BN) parameter learning from small data sets. First, the parameters in the network are analyzed and classified into classes as follows: parameters referring to different child states sharing the same parent configuration state and parameters referring to different parent configuration states sharing the same child state. Then, a novel beta distribution approximation based Bayesian estimation method is proposed, which is suitable for the learning of the first category parameters. Mean-while, previously proposed isotonic regression estimation method is employed to compute the second category parameters. Finally, simulations demonstrate the effectiveness of the proposed algorithm on improving the precision of Bayesian network parameter learning from small data set.