Dataset Quality Improvement for Fine-Grained Just-in-Time Software Defect Prediction

Recently developed fine-grained JIT-SDP models separately predict whether a changed file in a commit will cause a defect in the future or not (in other words, defect-inducingness), in contrast to traditional JIT-SDP models that only predict commits. Fine-grained JIT-SDP models also cost-effectively reduce the risk of overlooking defect-inducing changes in effort-aware JIT-SDP models by allowing developers to review only defect-inducing changed files in a commit. But the fact is that building machine learning models is a data-dependent process, so the quality of the data is crucial. Low data quality negatively affects the predictive performance, interpretability, and scalability of machine learning models. In the context of JIT-SDP, there is no study in the literature that directly focuses on data quality. In this light of information, we proposed a novel data quality improvement method for fine-grained JIT-SDP models considering software domain. We then demonstrated that our data quality improvement method increases predictive performance for within-project and cross-project fine-grained JIT-SDP models. In doing so, we open the door to JIT-SDP models that have good predictive performance, cost-effectiveness, and a low probability of overlooking project components that cause defects.