Defect Prediction Based on the Characteristics of Multilayer Structure of Software Network

Abstract

Software defect prediction can help us identify software defect modules and improve software quality. The existing defect prediction mainly analyzes the software code or the development process and uses the statistical feature data on the files or categories related to the software defects as the metrics. The method disregards the macroscopic integrity of software programs and the relevance of local defects to the surrounding program elements. For this reason, this study introduces a complex network technology into defect prediction, establishes a software network model, uses a complex network metric to design a set of metrics that can reflect the local and global features of defects, and proposes a dynamic prediction model optimization method based on the threshold filter algorithm. The effectiveness of the proposed metric and method is verified through comparison with the Predictive Model in Software Engineering dataset experiment and a practical engineering software data prediction experiment.