To address the challenges of high-dimensional features and imbalanced data in financial fraud detection, this paper proposes a fusion framework combining topological data analysis with an improved algorithm (GWOXGBoost). Based on the topological structure of corporate association networks, a dynamic relationship graph is constructed to establish a financial fraud detection model. The Grey Wolf Optimization Algorithm (GWO) and XGBoost technology, based on gradient boosting algorithms, are introduced to optimize model hyperparameters, enhance the model’s global learning capability and speed, thereby improving the effectiveness of financial fraud detection. Research findings indicate that the correlation coefficients of all 15 indicators are less than 0.00, with P < 0.01, demonstrating that the selected feature indicators effectively improve the quality of the financial fraud detection model. The model achieves an AUC value of 0.940, and among the two-class data processing methods, five indicators outperform the comparison model. The consistency between financial fraud detection results and actual conditions reaches a maximum of 97.26%.
