Towards Explainable Software Defect Prediction Models to Support SQA Planning

This thesis aims to tackle the explainability challenges of software defect prediction models. It first (1) investigates the impact that correlated metrics have on the explanations of defect prediction models; (2) identifies the best-automated feature selection techniques to mitigate correlated metrics for generating the explanations of defect prediction models; and (3) proposes the best model-agnostic techniques to explain the predictions of defect prediction models and generate actionable guidance to support SQA planning on what developers should do and should not do to prevent defects in the future.