CNN–BiLSTM-Based Framework for Dam Failure Risk Prediction Using Hydrometeorological Time-Series Data

Dam failure represents a critical hazard to human safety, infrastructure, and environmental stability, particularly under increasing climate variability and extreme hydrometeorological conditions. Accurate and reliable early warning systems therefore require predictive models capable of effectively capturing complex temporal dynamics in hydrometeorological processes. This study presents a CNN–BiLSTM-based deep learning framework for dam failure risk prediction using exclusively hydrometeorological time-series data. The proposed framework combines one-dimensional Convolutional Neural Networks (CNNs) for local temporal feature extraction with Bidirectional Long Short-Term Memory (BiLSTM) networks to model bidirectional temporal dependencies inherent in sequential data. The model was evaluated on a real-world hydrometeorological dataset consisting of 29,304 observations with pronounced class imbalance, where failure-related events account for approximately 10% of the samples. Model performance was assessed using evaluation metrics appropriate for imbalanced classification, including precision, recall, F1-score, receiver operating characteristic area under the curve (ROC AUC), and precision–recall area under the curve (PR AUC). The experimental results indicate that the proposed CNN–BiLSTM model achieves strong predictive performance, attaining an overall accuracy of 98.11%, a recall of 85.43% for failure events, and a PR AUC of 0.9591, while maintaining a low false alarm rate of 0.38%. These results demonstrate the capability of the proposed approach to balance early detection of potential failure events with operational reliability. The findings suggest that the CNN–BiLSTM framework offers a robust and data-driven solution for dam failure risk prediction and has significant potential for integration into real-time hydrometeorological monitoring and early warning systems.