An Explainable AI-Driven Hybrid Deep Learning Model for Secure Credit Risk Assessment and Real-Time Fraud Detection in Digital Banking

Abstract

Artificial intelligence (AI) ushered in the most revolutionary technological shift in human history. The financial sector may take advantage of this to improve customer experience, guarantee consumer protection, democratize financial services, and boost risk management. Although there is no shortage of modern machine learning models, it has proven difficult to develop and deploy systems that can withstand the rigors of real-world financial applications. This is mainly because these models lack the transparency and explainability that are necessary for constructing reliable technology. What makes this research unique is that it uses these worries about transparency to inform the creation of credit risk management policies through the use of an explainable AI (XAI) model. The suggested methodology can help regulators and financial institutions shape data-driven regulations, ensure fairness, and enhance trust by delivering a comprehensive knowledge of the fundamental causes impacting AI forecasts. The purpose of this research is to provide a transparent artificial intelligence model for managing credit risk, with a focus on elucidating the dangers of P2P lending. Using important explanatory factors as inputs, the model uses Shapley values to produce AI predictions. The two models with the best accuracy rates were the random forest and decision tree ones, with 0.98 and 0.96 points, respectively. Decision tree and random forest models achieved accuracies of 0.99 and 0.98, respectively, when evaluated on a bigger dataset; the model's accuracy levels were steady throughout. The problem's binary classification led to the selection of these models for explainable AI (XAI) modeling. The XAI models were presented as global and local surrogates using LIME and SHAP.