Implementing Explainable AI for Proactive Regulatory Compliance and Auditing in Financial Markets

A credit risk is the possibility of monetary loss due to the borrower's partial or whole failure to fulfil their obligations. This manuscript aims to examine XAI for pioneering feasible regulatory compliance and auditing in the credit risk assessment and loan performance analysis of the financial industry. The methodology integrates machine learning models, including RF and GB, to evaluate credit risk through a dataset comprising 887,379 entries across 74 columns. The study under discussion also focuses on data preprocessing required for correct label encoding and the division of data into training and testing sets. Evaluation metrics, including the ROC curve, accuracy, precision, recall, and F1score, are utilized to assess the model. This means that Gradient Boosting is more accurate and probably the best model, with an accuracy of 98.56% and ROC-AUC of 98.26%, while Random Forest has high precision though low recall. In addition, LIME is the XAI approach used in the research to improve the credit risk model's interpretability. The goal of this work is to deliver usable and pro-active methods for machine learning interpretation in order to assist financial institutions with regulatory compliance and mitigating risks and uncertainty in the financial environment effectively.