Survey of Data-Driven Approaches for Credit Risk Evaluation in Banking Using AI Techniques

Abstract

Credit risk measurement is one of the most prominent problems in modern banking and financial services that has a direct impact on the lending operations, profitability, and stability. Conventionally, qualitative, experience-based methods have been applied in assessing credit risk based on transactional jeopardy, inherent peril and concentration risk in loan portfolios. Nonlinear patterns and dynamic risk behavior are however usually not well captured using the traditional methods, particularly due to increasing complexity of the financial markets, as well as due to the profile of diverse borrowers, and the presence of large volumes of data. Credit risk assessment has been transformed by recent advancements in AI and ML, which have made it data-intensive, adaptable, and scalable. When dealing with high-dimensional, unbalanced, and heterogeneous financial data, the approaches that have proven to be effective include logistic regression, random forests, gradient boosting machines, and deep neural networks. Additionally, alternative sources of data, such as transactional, demographic, behavioral, and social data can be utilized to predict credit scoring with increased accuracy and reduced bias with the aid of advanced preprocessing and feature engineering. In addition to risk evaluation, AI is also used to identify fraud, predictive analytics, regulatory compliance, and chatbots. Although the above advantages are there, there are still issues of data privacy, ethics and model interpretability. The paper indicates traditional and AI-based methods of credit risk assessment.