A Method-Level Defect Prediction Approach Based on Structural Features of Method-Calling Network

Software defect prediction models help testers find program modules that have a high probability of having defects. A method-calling network can express the dependencies between methods in program. Existing approaches do not sufficiently utilize to characterize structural features methods. To address this problem, study, it is proposed for first time characteristics are obtained by analyzing network, and new approach at method-level. Specifically was constructed metrics were obtained. Next, embedding technique (node2vec) used automatically encode structure into low-dimensional vector obtain metrics. Finally, they combined with code construct models. We evaluated our on 13 open-source software systems. The experimental results show method improved values area under receiver operating characteristic curve 2.5% 6.7% Matthews correlation coefficient 13% 178.4% compared baselines. Therefore contains rich methods, extend method-level further improve performance