A Type-2 Fuzzy Logic Approach to Explainable AI for regulatory compliance, fair customer outcomes and market stability in the Global Financial Sector
Janet Adams Business Banking TSB Bank London, UK
Hani Hagras The Computational Intelligence Centre, School of Computer Science and Electronic Engineering University of Essex, Colchester, UK
Abstract—The field of Artificial Intelligence (AI) is enjoying unprecedented success and is dramatically transforming the landscape of the financial services industry. However, there is a strong need to develop an accountability and explainability framework for AI in financial services, based on a risk-based assessment of appropriate explainability levels and techniques by use case and domain.
This paper proposes a risk management framework for the implementation of AI in banking with consideration of explainability and outlines the implementation requirements to enable AI to achieve positive outcomes for financial institutions and the customers, markets and societies they serve. The work presents the evaluation of three algorithmic approaches (Neural Networks, Logistic Regression and Type 2 Fuzzy Logic with evolutionary optimisation) for nine banking use cases. We review the emerging regulatory and industry guidance on ethical and safe adoption of AI from key markets worldwide and compare leading AI explainability techniques.
We will show that the Type-2 Fuzzy Logic models deliver very good performance which is comparable to or lagging marginally behind the Neural Network models in terms of accuracy, but outperform all models for explainability, thus they are recommended as a suitable machine learning approach for use cases in financial services from an explainability perspective.
This research is important for several reasons:
there is limited knowledge and understanding of the potential for Type-2 Fuzzy Logic as a highly adaptable, high performing, explainable AI technique;
there is limited cross discipline understanding between financial services and AI expertise and this work aims to bridge that gap;
regulatory thinking is evolving with limited guidance worldwide and this work aims to support that thinking;
it is important that banks retain customer trust and maintain market stability as adoption of AI increases.
Read the full paper below