Dynamic Risk Analysis Model in Digital Financial Systems Using Deep Reinforcement Learning

Abstract

Digital transformation in the financial system has created new risk dynamics that are more complex and unpredictable with conventional methods. Digital asset market volatility, cyberattacks, and changes in user behavior demand the development of new approaches in risk management. This study aims to develop a dynamic risk analysis model using Deep Reinforcement Learning (DRL), especially with the Deep Recurrent Q-Network (DRQN) approach, to improve the effectiveness of real-time risk detection and mitigation in the digital financial system. The research method used is an exploratory-descriptive qualitative approach with data obtained through documentation studies, field observations, and digital transaction simulations. The DRQN model is compared to the DQN model to evaluate the accuracy, speed of response, and effectiveness of adaptation to changing transaction patterns. The results showed that DRQN was able to increase the accuracy of risk prediction by up to 91.4% and accelerate the response time to an average of 1.8 seconds. In addition, the model has proven to be more adaptive to sequential data and rapid market fluctuations. The conclusion of this study confirms that the integration of DRLs in the digital financial system can increase the resilience of the system to complex risk dynamics. The practical implication of this research is to encourage the financial industry and regulators to adopt adaptive learning-based technologies in risk mitigation strategies in the digital economy era.