Software Defect Prediction Using Process Metrics Systematic Literature Review: Dataset and Granularity Level

Software defect prediction is an important part to ensure software quality and increase cost-efficiency. The core aspect prior to software defect prediction is software metrics. More research of process metrics existed and rapidly developed as the fact that process metrics can outperform code metrics. The development of process metrics is diverse aligning with various process metrics generated in the different publication using different dataset and granularity level, thus the development state of process metrics need to be further tracked. Recent research on software defect prediction using process metrics has been carried which complements or argues another studied related to the implementation of process metrics, dataset, and granularity level. This literature review has collected 17 literature studies that aim to depict the current research used of software defect prediction applying software process metrics, dataset, and granularity. Seven commonly used process metrics are identified such as NDC, NML, ADEV, EXP, LA, LD, and COMM. Process metrics development has been implemented in the previous research such as the application of aggregated process metrics and commits detailed information metrics. The selected research shows that 87% of research are using a public dataset while the rest are using a combined dataset. This result depicts the limited accessibility of private and industrial dataset for defect prediction. Following analysis of software granularity presents 4 levels of granularity used in selected research such as a functional, class, file, and features level. File-level defect predictions are mostly used for defect prediction in selected research and prove good defect prediction performance. The results of this literature review also identify the proposed studies which is software defect prediction using developed process metrics in the different datasets and features-level granularity to strengthen the previous research.