This study constructs a machine learning-based risk identification and warning model for financial risks in the digital creative industry. By collecting financial data from A-share listed companies in the digital creative industry between 2018 and 2023, and performing data cleaning and feature engineering, an indicator system containing 78 features was developed. A hybrid XGBoost and LSTM architecture was used to design the warning model, combining static financial features with dynamic industry chain risks. Experimental results show that the model achieves an accuracy of 83.7% and can issue warnings 2.3 quarters in advance, significantly outperforming traditional methods. SHAP value analysis indicates that R&D investment intensity, intellectual property value, and market share are the most valuable features for warning signals. The model demonstrates good stability and interpretability in practical applications, providing an effective tool for risk management.