Bayesian Network Based Program Dependence Graph for Fault Localization

Xiao Yu, Jin Liu, Zijiang James Yang, Xiao Liu, Xiaofei Yin, Shijie Yi

Abstract: Some probabilistic graphical models such as probabilistic program dependence graph (PPDG) have been used in fault localization. However, PPDG is insufficient to reason across nonadjacent nodes and only support making inference about local anomaly. In this paper, we propose a novel probabilistic graphical model called Bayesian Network based Program Dependence Graph (BNPDG) that has the excellent inference capability for reasoning across nonadjacent nodes. We focus on applying the BNPDG to fault localization. Compared with the PPDG, our BNPDG-based fault localization approach overcomes the reasoning limitation across nonadjacent nodes and provides more precise fault localization by taking its output nodes as the common conditions to calculate the conditional probability of each non-output node. Experiment results show that our BNPDG-based fault localization approach outperforms its rivals.