Discrete Bayesian network parameter learning based on monotonic constraint

With respect to the problem of learning parameters of discrete Bayesian network from small sample data, a parameter learning algorithm is proposed based on the monotonic constraint. Firstly, the mathematical model of the monotonic constraint is built to express the qualitative prior information. Then, the monotonic constraint is integrated into the Bayesian estimation as Dirichlet prior and the modified Bayesian estimation is employed to learn parameters. Finally, the proposed algorithm is compared with maximum likelihood estimation and isotonic regression by simulation experiments. The experimental results show that the proposed algorithm is better than maximum likelihood estimation and isotonic regression on accuracy, and its' timeliness is between the two algorithms.