Software defect prediction using hybrid techniques: a systematic literature review

Software defect prediction is the process of developing predictive models that helps in the early identification of defect-prone modules based on software metrics and defect data. It enables the project managers to allocate resources optimally. Recently search-based techniques have been widely adopted for providing optimal solutions to develop effective software defect prediction models. When merged with statistical or machine learning techniques, these techniques give rise to hybridization. Hybrid techniques significantly improve the performance of search-based techniques and the quality of solutions produced by them. In this systematic review, we have investigated 72 papers published from January 2000 to December 2021 that ascertain the use of hybrid techniques and their effectiveness in predicting software defects. The review identifies and evaluates various techniques, experimental design, fitness functions, performance measures, threats to validity, and well-known statistical tests in the studies that apply these hybrid techniques. We also compared the performance of hybrid techniques among themselves and statistical and machine learning techniques present in the literature. According to the findings, the predictive capability of hybrid techniques is encouraging. The results of this review paper will serve as guidelines for the researchers and practitioners, allowing them to make appropriate choices for applying hybrid techniques in the software defect prediction process.