Abstract
Recently, software analytics began to explain the reasons behind the predictions of machine learning (ML) and artificial intelligence (AI) models for various aspects of software projects using model-agnostic techniques derived from explainable artificial intelligence (XAI) domain. However, there is no guarantee that different modelagnostic techniques generate consistent explanations for the same predictions. Therefore, practitioners may obtain different insights depending on the technique they use. This article discusses in detail the problems caused by the inconsistent explanations generated from different model-agnostic techniques. In addition, we propose a method to integrate inconsistent explanations to derive information which can provide more useful and reliable software analytics to practitioners. Lastly, we discuss the future research directions for which explainable artificial intelligence for software engineering (XAI4SE) community to carry forward in exploiting the practical value of model-agnostic techniques.
Published Version
Talk to us
Join us for a 30 min session where you can share your feedback and ask us any queries you have
Schedule a call