Credit Risk Assessment Using Reinforcement Learning and Graph Analytics on AWS
Digital Banking, JP Morgan Chase, USA.
Review Article
World Journal of Advanced Research and Reviews, 2023, 20(01), 1399-1409
Publication history:
Received on 02 September 2023; revised on 20 October 2023; accepted on 27 October 2023
Abstract:
Traditional credit risk assessment models rely on static scoring mechanisms that fail to capture dynamic borrower behavior patterns and interconnected financial relationships, resulting in suboptimal lending decisions and increased default rates particularly during economic volatility. This paper presents an innovative intelligent credit risk assessment framework combining Deep Reinforcement Learning with Graph Neural Networks, deployed on Amazon Web Services infrastructure for real-time adaptive decision making. Our approach integrates AWS Neptune for graph database management, Amazon Personalize for behavioral modeling, and AWS SageMaker for distributed reinforcement learning training across heterogeneous financial datasets. The system employs a novel Multi-Agent Deep Q-Network architecture that learns optimal lending strategies through interaction with simulated economic environments, while Graph Attention Networks model borrower interconnectivity and systemic risk propagation. Advanced feature engineering incorporates temporal transaction patterns, social network analysis, alternative credit data sources, and macroeconomic indicators processed through Amazon Timestream and AWS Glue. Experimental evaluation using 4.2 million loan applications from 78 financial institutions demonstrates 87.3% accuracy in default prediction with 34% reduction in false positives compared to traditional FICO-based models. The reinforcement learning agent achieved 23% improvement in portfolio return-on-investment while maintaining regulatory compliance through AWS Config and automated policy enforcement. Integration with Amazon Fraud Detector enables real-time anomaly detection, while AWS Lambda provides sub-200ms credit decisions with automatic scaling capabilities. This research addresses the critical gap between static risk models and dynamic market conditions, providing an adaptive, scalable, and regulatory-compliant solution for next-generation credit risk management in digital banking environments.
Keywords:
Credit Risk Assessment; Reinforcement Learning; Graph Neural Networks; AWS Neptune; Behavioral Analytics; Adaptive Lending; Systemic Risk
Full text article in PDF:
Copyright information:
Copyright © 2023 Author(s) retain the copyright of this article. This article is published under the terms of the Creative Commons Attribution Liscense 4.0