Designing a feature selection method based on explainable artificial intelligence

Abstract Nowadays, artificial intelligence (AI) systems make predictions in numerous high stakes domains, including credit-risk assessment and medical diagnostics. Consequently, AI increasingly affect humans, yet many state-of-the-art lack transparency thus, deny the individual’s “right to explanation”. As a remedy, researchers practitioners have developed explainable AI, which provides reasoning on how infer individual predictions. However, with recent legal initiatives demanding comprehensive explainability throughout (development of an) system, we argue that pre-processing stage has been unjustifiably neglected should receive greater attention current efforts establish explainability. In this paper, focus introducing an integral part stage: feature selection. Specifically, build upon design science research develop framework for We instantiate running software artifact evaluate it two group sessions. Our helps organizations persuasively justify selection stakeholders and, comply upcoming legislation. further provide consisting meta-requirements principles