A Case Study of Defect Prediction¶
In today’s increasingly digitalized world, software defects are widespread and enormously expensive, but they are very hard to detect, predict, and prevent. Thus, a failure to eliminate software defects in safety-critical systems could result in serious injury to people, threats to life, death, and disasters. Here, we present 3 use-case scenarios of eXplainable AI defect prediction framework to achieve the followings:
Generating fine-grained predictions in order to help developers localize which lines of code are the most risky so developers can allocate limited software quality assurance activities in a cost-effective manner[WTT+20].
Generating explanations for each prediction to help developers understand why a file is predicted as defective[JTDG20b].
Generating actionable guidance to help managers chart appropriate quality improvement plans.