Comparative analysis of machine learning models with SHAP interpretation for causes of highway flood-damage blocking.

Highway flood-damage blocking poses a critical threat to transportation system resilience, yet risk assessment is often limited by insufficient modeling of temporal disaster evolution, highly imbalanced data, and weak model interpretability. To address these challenges, this study proposes an integrated modeling framework that combines temporal data augmentation, machine learning, and interpretable mechanism analysis. Three data-balancing strategies-Time-series Generative Adversarial Network (TimeGAN) augmentation, undersampling, and a hybrid approach-were systematically compared to handle imbalanced temporal data. Six machine learning models (Logistic Regression, Support Vector Machine, Random Forest, Decision Tree, eXtreme Gradient Boosting, and Multilayer Perceptron) were evaluated, and SHapley Additive exPlanations (SHAP) was used to quantify factor contributions and explore nonlinear effects and interaction patterns. Results show that the Multilayer Perceptron trained on TimeGAN-augmented sequences achieved the highest performance, with an F1 score of 49.81% and PR-AUC of 49.46%. SHAP analysis identified key drivers and their threshold effects: Daily precipitation exceeding 2.8 mm, 7-day effective precipitation (EP 7) exceeding 22 mm, temperature above 21 °C, and average road-stream distance within 1 km (ARSD) above 0.15 km significantly increase the risk of highway flood-damage blocking. High temperature conditions are more likely to coincide with heavy precipitation and elevated EP 7, and their combined effects further amplify blocking risk. Factor contributions also varied across methods, reflecting SHAP's ability to capture nonlinear effects and reveal interaction patterns, whereas linear regression mainly reflects independent linear effects. By integrating temporal generation, systematic model evaluation, and interpretable analysis, this study enhances the accuracy and reliability of highway flood-damage blocking prediction, providing quantitative guidance for flood damage prevention and resilience improvement of highway systems.