Combining MDL and BIC to Build BNs for System Reliability Modeling

Bayesian networks (BNs) is widely used for system reliability modeling because of its versatility. It is crucial to build BNs from reliability data without experts' intervene. However, the BNs structure learning is still an open problem. Traditional approaches always integrate a structure scoring metric with a particular heuristically searching method. This usually overfits or underfits the data. In this paper, we propose a new method that combines the Minimal Description Length principle (MDL) and Bayesian Information Criterion (BIC) to partially overcome the weakness of traditional mothods. Simula-tions show that the proposed method can effectively reduce the overfitting and underfitting of the BN model, thus improving the accuracy of the reliability estimation results.