Explainable AI framework for the financial rating models
Se Bin Kim, Jihwan Woo · 2021
As artificial intelligence has developed and applied over many industries, demands for explaining the results have been increased respectably. Recently EU created a new privacy regulation to guarantee the right of customers to have an explanation on the prediction model. Due to increasing demands on transparency for AI model, the techniques of explaining artificial intelligence (XAI) have been the center of attention. Heavy regularized industries, such as finance, have introduced many artificial intelligence powered software and models, and have faced huge pressures as well as demands to provide enough explanations from both customers and the authorities. There have been many XAI algorithms to interpret the model as well as the outcome, but there is little work done to explain the outcomes in comparison. In real world, it is often demanded to explain the reasons for alterations by comparison against previous ones such as changing credit rates over time which can be an essential factor for an individual’s life. In this paper, we suggest a comparison explanation framework using two outputs which can be financial ratings [11]. We have demonstrated the explanation framework on real world data with various ML algorithms and local surrogate models. We evaluate the quality of the explanation with the domain expert’s survey. The suggested framework received positive feedbacks from the experts and proves that it may applied to domain specialized areas. Also, the research contributes an alternative view of XAI in terms of practical usage at the field of finance.