Machine learning approach for software defect prediction using multi-core parallel computing

摘要

Defect prediction in software development is a very active topic of study. Software defect prediction (SDP) findings give the list of defect-prone source code artefacts, enabling quality assurance teams to efficiently allocate limited resources for validating software products. In order to enable both developers and reduce the time to market for more dependable software products, software defect prediction tools will play an increasingly significant role. Many machine learning approaches are present in the existing literature for SDP to enhance the performance of the software development team. However, very little work is reported for SDP using multi-core parallel computing. In this paper, a multi-core parallel machine learning approach for software defect prediction is proposed to classify a component as defective or non-defective. The proposed model has been built, trained and tested by varying the number of CPU cores involved in the processing. Extensive empirical studies have been conducted by applying the proposed approach on 11 software systems of NASA/PROMISE and other relevant repositories. The proposed approach has been compared with various state-of-art machine learning models to investigate the proposed models' supremacy in comparison with the other existing models. The experimental results indicate that the predictive performance of the proposed model is improved, and execution time is decreased by involving a greater number of CPU cores. Through evaluation of calculated results, it has been observed that the multi-core parallel processing Random Forest approach gives the best predicting performance parameters values nearly 99 or 100%. Moreover, the proposed approach performs significantly better in accuracy, precision, recall, F-Measures, and AUC compared to other machine learning models.