Explainable Artificial Intelligence (XAI): Conception, Visualization and Assessment Approaches Towards Amenable XAI

Abstract

AbstractFor last decade, due to the accessibility of huge databases and recent advancements in deep learning methodology, machine learning systems have arrived at or transcended tremendous performance in a spacious variety of tasks. One can see this speedy development in speech analysis, image recognition, sentiment analysis, strategic game planning and many more, for e.g. in medical field, it’s used for diagnosing different diseases, like breast cancer etc., based on their symptoms. But many state-of- the-art models is facing lack of transparency and interpretability which is a major hindrance in many applications, e.g. finance and healthcare where visualization, interpretation and explanation for model's decision is an obligation for trust. This is an implicit problem of the current techniques carried by sub-symbolism (e.g. Deep Neural Networks) that were not shown in the last hype of AI (specifically, rule based models and expert systems). Models underlying this problem come within the so-called Explainable AI (XAI) field, which is extensively acknowledged as a racial feature for the practical deployment of AI models. As a result, explainable artificial intelligence (XAI) has turned into scientific interest in last recent years. So, this chapter epitomizes contemporary developments in Explainable AI that describes explainability in Machine Learning, constituting a fiction definition of explainable Machine Learning that envelopes such prior conceptual propositions with a considerable focus on the audience for which the explainability is needed. Except of this definition, this chapter starts a confabulation on its various techniques that are essentials for analysing interpretability and explainability of Artificial Intelligence, and also gives a comparison between two medical experiments, that are based on predicting heart disease using disparate Explainable Artificial Intelligence techniques, which can give a lead for researchers as well as practitioners or newcomers in the field of Artificial Intelligence for selecting suitable methods with Explainable AI to grasp the advances of AI in their action sectors, without any previous bias for its dearth of interpretability.KeywordsExplainable artificial intelligenceMachine learningInterpretabilityExplainabilityBlack box modelPost-Hoc method