A Data-Driven Approach to Dam Infrastructure Monitoring: Enhancing Prediction Accuracy by Systematic Rainfall Event Classification

Abstract The safety of aging dam infrastructure has emerged as a critical issue as many social infrastructure systems experience functional degradation over time. While numerous monitoring devices have been deployed to evaluate dam conditions, the reliability of measurement data is often compromised due to harsh environmental factors, missing values, and abnormal fluctuations. These challenges highlight the necessity of ensuring data quality through both accurate point-level validation and comprehensive pattern analysis. Existing approaches often fail to systematically distinguish between environmental conditions such as rainfall and non-rainfall events, limiting the predictive reliability of artificial intelligence (AI) models in dam safety applications. To address this limitation, this study proposes a robust prediction framework that combines targeted data preprocessing with an eXtreme Gradient Boosting (XGBoost) model to enhance the reliability of dam measurement forecasts. Measurement data from infiltration turbidity and leakage meters were categorized based on rainfall occurrence to reflect environmental variability. The preprocessing method emphasizes structured training data reconstruction rather than the simple removal of outliers by identifying and separating time segments influenced by rainfall events. The proposed approach allows the model to better capture complex nonlinear patterns in dam behavior and improves prediction accuracy across various dam conditions. This study offers a practical contribution to the development of high-reliability forecasting systems for dam monitoring and lays the groundwork for future adaptive dam safety management strategies.