Using sum product networks to predict defects in software systems

Software defect prediction techniques are utilized in the early stages of the software development process to reduce costs and save time on corrective maintenance activities. While many approaches based on machine learning techniques have been proposed to predict bugs in software systems, the Sum Product Network has not yet been evaluated in a bug prediction scenario based on software metrics. Therefore, this paper proposes an approach based on SPNs to predict defects in software systems. The proposed approach is a classifier based on a SPN, learned from historical data of previous projects, and then used to predict faulty modules in new projects. The proposed approach is evaluated on 21 projects from four well-known datasets (NASA, Promise, Relink, and AEEEM). The obtained results are compared against eight Software defect prediction baseline methods. The empirical results show that DPSPN has the potential to effectively predict defects in software systems.