Financial fraud detection model based on dual-layer knowledge graph

Corporate financial fraud is becoming increasingly complex and covert, necessitating the development of intelligent identification frameworks that combine accuracy and interpretability. This study proposes a detection method based on a two-layer knowledge graph. A two-layer structure is constructed, combining cross-layer path mining and rule vector discrimination to achieve end-to-end closed-loop detection. Experiments show that the method maintains stable performance as the enterprise scales up. In a sample of 700 enterprises, the method achieves an accuracy of 94.2%, a precision of 92.7%, a recall of 93.5%, an F1 score of 93.1%, an AUC of 0.95, and an MCC of 0.88. In a multi-enterprise scenario in the manufacturing industry, the detection accuracy reaches a maximum of 97%, AUC 0.975, and rule trigger rate 87%. After 100 iterations of rule mining, approximately 390 high-quality rules are retained, with an average support exceeding 0.069. This method demonstrates significant advantages and provides a practical intelligent framework for enterprise auditing and supervision.