Untangling Explainable AI in Applicative Domains: Taxonomy, Tools, and Open Challenges

Recently, a paradigm shift is observed toward Industry 5.0, where tasks (processes) are automated at massive scales. This shift has initiated modern developments in artificial intelligence (AI) to support a plethora of applications like manufacturing, health care, vehicular net- works, and others. However, owing to the black-box nature of AI models, the research has shifted toward the proposal of novel techniques that aim toward the explainability and validity of these AI models. Thus, explainable AI (XAI) has become a norm in modern applicative domains, and the study of its frameworks and tools has become the buzzword among researchers. Thus, the paper intends to present the key concepts of XAI and aims at improving the model transparency. The survey systematically untangles the key concepts of XAI and presents a solution taxonomy in different applications. Modern XAI techniques are classified as self-explanatory, visual-based-model-agnostic, global surrogate, and local surrogate-model-agnostic. We also cover the tools and frameworks of XAI and discuss the open issues and challenges in practical realization. Thus, the survey intends to arm AI practitioners to design optimal solutions to realize XAI in practical use-case setups.