Optimizing Credit Scoring with MLOPS: Enhancing Risk Management and Profitability

This study presents the implementation of an MLOps-driven framework to optimize credit scoring models by integrating MLflow for model tracking, version control, and deployment, alongside automated performance monitoring using Airflow. The proposed approach replaces a manually maintained credit risk assessment pipeline with a structured and reproducible system that enables semi-automated model retraining and controlled updates. Key enhancements include automated feature aggregation with PostgreSQL, real-time model monitoring, and structured governance via MLflow Registry, ensuring model transparency, traceability, and compliance with financial regulations. The transition from manual to automated processes has led to a 91.7% reduction in model development time, an 85–95% acceleration in model update cycles, and a 6.67% improvement in AUC-ROC with a 40% increase in the Gini coefficient. The study demonstrates how MLOps best practices improve credit risk management by minimizing human intervention in model maintenance while maintaining oversight through manual validation before production deployment. These findings provide a structured framework for financial institutions seeking to enhance scalability, decision-making efficiency, and regulatory compliance in credit scoring. Keywords: MLOps, credit scoring, MLflow, risk management, model monitoring, automation, feature engineering, compliance.