Abstract Urban areas are increasingly affected by intense rainstorm-induced flooding, posing serious risks to human life and property. To reduce the impact of such disasters and promote urban safety and sustainable development, a systematic assessment of Urban Flood Disaster Risk (UFDR), along with appropriate management strategies, is essential. Due to the high cost and complexity of acquiring and processing numerous potential indicators, identifying the most influential predictive variables is critical. This study integrates adaptive fuzzy logic with machine learning techniques to predict flood probabilities and develop evidence-based mitigation protocols. The proposed framework incorporates 13 carefully selected evaluation indicators, categorized into three dimensions: hazard triggers, environmental susceptibility, and community vulnerability. Indicator weights are determined through a combination of subjective (Analytic Hierarchy Process) and objective (CRITIC) weighting methods. Dynamic risk assessment is conducted using the Variable Fuzzy Pattern Evaluation (VFPE) model, while temporal features are automatically extracted using one-dimensional convolutional neural networks (1D-CNN). Flood probability is predicted using several machine learning algorithms, including Random Forest (RF), Decision Tree (DT), Extreme Gradient Boosting (XGBoost), and Support Vector Machine (SVM). The contribution of each input variable is assessed using feature importance scores derived from the RF model, averaged over a leave-one-out cross-validation (LOOCV) process. Results indicate that the SVM model achieves the highest accuracy and reliability for the multi-class classification task, particularly in identifying high-risk events. RF and XGBoost also demonstrate strong performance, offering a balance between predictive accuracy and model interpretability. Overall, the proposed methodology provides an effective and data-driven approach to support urban flood risk assessment and disaster mitigation planning.
Sun et al. (Thu,) studied this question.