Integration of Audit Rules and Data-Driven Risk Identification Algorithm Models

Abstract

With the rapid development of information technology, data-driven risk identification has become widely applied in the auditing field. However, existing methods often overlook the integration of audit rules, limiting their performance in complex scenarios. This paper proposes a model that integrates audit rules with data-driven risk identification, where the audit rules fusion module and deep neural networks collaborate to achieve a deep fusion of rule knowledge and data features. Experimental results show that, on a custom dataset, our approach significantly outperforms baseline models in accuracy, recall, F1 score, and AUC, with accuracy reaching 89.3%, improving by 1.2% over the optimal deep neural network and F1 score improving by 2.3%. Even with 50% noise intensity, the model maintains an F1 score of 81.3%, demonstrating strong noise robustness. Cross-domain dataset tests validate its generalization ability. Ablation studies reveal that both the audit rules fusion module and the denoising mechanism are crucial for model performance improvement, with their combined effect reducing the false positive rate by approximately 8.6%. This study provides a feasible path for integrating audit rules with data-driven methods and offers a new solution for risk identification in complex audit scenarios.