With the rapid development of information technology, the audit work of accounting firms is gradually transitioning towards informatization and intelligence. However, the level of informatization construction among these firms is uneven, and the high-dimensional characteristics and the complexity of abnormal distribution of audit data pose significant challenges to audit work. To improve the efficiency and accuracy of anomaly detection, a hybrid model based on the fusion of clustering and classification is proposed in this study. The results show that the hybrid model achieves precision, recall, and F1 scores of 96.1%, 93.2%, and 94.6% respectively on real datasets, significantly outperforming single algorithms. Meanwhile, in simulated datasets with a low proportion of anomalies, the F1 score of the hybrid model remains at 91.6%, verifying its robustness across different scenarios. Furthermore, the system’s false alarm rate is controlled below 0.63%, demonstrating exceptional anomaly detection capabilities. In conclusion, this study, combining algorithm optimization with practical application validation, provides a new technical pathway for firms to enhance their audit informatization level and lays the foundation for intelligent auditing in complex data scenarios.