AI-Driven Credit Scoring Models: Enhancing  Accuracy and Fairness with Explainable Machine Learning

Abstract—Credit scoring remains a crucial challenge for financial institutions, remarkably when rapid, high-volume evaluations are required. This study presents an AI-driven, real-time credit scoring system using LightGBM integrated with a high-performance data pipeline. The system can process up to 9,500 transactions per second with minimal latency of 8 milliseconds. Addressing the data imbalance in credit scoring datasets, the model also ensures transparency through Explainable AI (XAI) techniques, specifically using SHAP values to interpret the model’s credit risk predictions. Experimental results on real world financial transaction datasets show that the proposed system achieves an accuracy of 98.7%, with precision and recall scores exceeding 94% and 91%, respectively. Compared to baseline models and other approaches in the literature, our system demonstrates superior scalability, processing speed, and accuracy. This solution offers a robust and efficient framework for real-time credit scoring, ensuring high performance, transparency, and low latency, making it ideal for modern financial applications. Index Terms—Federated Learning, Explainable AI, Fraud Detection, Data Privacy, Imbalanced Datasets, Financial Institutions, Model Transparency, Collaborative Learning, Customer Confidentiality, Risk Management