Explainable AI for Credit Scoring in Cloud Enterprise Banking Systems

In banking, traditional credit scoring models are based on the use of machine learning models that can be quite complex, and generally act as a black box, which means that financial institutions are not able to explain credit decisions to the regulators or the customers, especially in cloud-based enterprise settings where transparency and accountability are both critical. This work will focus on the design and testing of a model XAI framework that could improve the interpretability of models whilst not compromising predictive performance of cloud enterprise banking systems in terms of credit scoring. We used a combined version of XAI methodology, LIME (Local Interpretable Model-agnostic explanations), SHAP (SHapley Additive exPlanations), and ad code feature attributions on$\mathbf{5 0, 0 0 0}$credit application datasets. This model was implemented on the AWS cloud and tested on traditional credit scoring models based on the metrics accuracy, precision, recall, and interpretability. The comparisons of performance were done against traditional logistic regression, random forest, and neural networks. The proposed XAI framework proved to be an interpretable model with 89.2 % accuracy in credit default prediction and with 12 % improvement on interpretability scores against the baseline models with comprehensive explanations to every decision made. The cloud architecture achieved 99.7 % uptime and a response time of 150 ms on average, which is suitable for deploying an enterprise-level system. This study found that explainable AI can effectively achieve a trade-off between predictive capability and transparency when used in credit scoring to help financial organizations achieve regulatory compliance implications to ensure a cloud environment without compromising the aspects of accuracy.