Modeling Audit Workflow Dynamics with Deep Q-Learning for Intelligent Decision-Making

Abstract

This study addresses the problem of insufficient adaptability in traditional audit programs under dynamic environments by proposing an adaptive adjustment framework based on deep Q-Learning. First, the audit workflow is modeled as a Markov decision process. State representations are constructed by combining account features, transaction behaviors, and internal control indicators. A reward function is designed to balance risk identification and resource optimization. Then, a deep neural network is used to approximate the optimal Q-value function. Experience replay and target network mechanisms are adopted to enhance model training stability and generalization ability. Through a series of comparative experiments, the proposed method is shown to outperform traditional reinforcement learning methods in terms of average reward, adjustment efficiency, and resource consumption control. In addition, hyperparameter sensitivity experiments are designed around optimizer selection, learning rate settings, and resource allocation strategies. These experiments further analyze the impact of key training parameters on model performance. The experimental results demonstrate that the proposed method effectively improves the flexibility and precision of audit program execution. It shows strong empirical results and practical application potential. This provides solid technical support for intelligent decision-making in intelligent audit systems.