Construction of Digital Financial Risk Prediction Model based on Artificial Intelligence

Background: To address complex financial market risks, this study proposes a fusion model combining XGBoost and LSTM for digital financial risk prediction. Methods: The model uses LSTM to process dynamic time series data and XGBoost for static structured data, integrating features through a dual-channel architecture. Results: Experimental results on CSI 300 Index data show the fusion model outperforms single models in classification accuracy (0.89), AUC value (0.92), and F1 score (0.85). It demonstrates superior robustness under noise interference, with smaller prediction errors and higher stability. Concluding Remarks: This approach effectively overcomes limitations of traditional models reliant on static features, offering enhanced predictive accuracy and generalization.