Learning Bayesian Networks Parameters via Monotonicity Constraints

Abstract

Learning Bayesian networks (BNs) parameter is a challenging task as it relies on a large amount of reliable and representative training data. However, the accuracy of parameter learning is critical. In this paper, we propose PML_margin method to enhance monotonicity to achieve better results. We apply Spearman rank correlation coefficient to detect monotonicity constraints and to adjust the margin of cumulative distribution. Experimental results on standard BNs show the effectiveness of the PML_margin, compared with the MLE, MAP and RHO_PML method, respectively.