XAI Techniques: Review of Last 5 Years Explainable Artificial Intelligence Techniques

Abstract

Due to availability of computational power and data, Machine Learning (ML) has gained lots of popularity in recent time. ML based solutions growing veryvastly, many industries already shown interest onit and using suchsolutions.ML solutions are significantly deployed in various key domains like healthcare, security, finance and has been used successfully with utmost human level of accuracy. With the availability of medical data, ML models are capable enough to identify the root cause of any ailment. Though ML models are producing high level of accuracy, still as a consumer of ML models no one is sure why the model take such decision, especially in case where ML models are based on Deep Neural Network (DNN), which is having very complex architecture.That’s the reason, ML models are also known as black-box model. For various industries like healthcare, security etc.merely relying on ML model perditioncan be hazardous. One should definitely want to know why such decision has been made by the model before taking further actions. That’s the reason, along with AI-ML research a new requirement is emerging in the market and hence lots of research is happening on explaining the decision made by model. Explainable AI (XAI) term is used for explaining the modeland identifying the reasons for its prediction. Lots of discussions and research is happening in converting a black-box model to white-box model (transparent model). Purpose of this paper is to explain what all concrete research work has been done in last 5 (2015-2019) years inXAI domain in terms of developing new approaches, methods, tools, frameworksusedfor explaining or interpreting ML models