Scholarlink

<p>Rapid urbanization has heightened pluvial flood risks in developing cities, yet most machine learning (ML) studies rely on single-model frameworks with limited comparative evaluation. This study develops a multi-model benchmarking framework for urban pluvial flood susceptibility assessment, integrating nine explanatory variables selected via variance inflation factor (VIF) analysis to reduce multicollinearity. Four ML algorithms—Support Vector Machine (SVM), Random Forest (RF), Gradient Boosting (GB), and eXtreme Gradient Boosting (XGB)—were tested with Bayesian Optimization (BO) and Particle Swarm Optimization (PSO) for hyperparameter tuning, using 200 balanced flood and non-flood samples from Wuxi, China. Key findings are as follows: (1) The PSO-XGB model achieved the highest predictive performance (test accuracy = 0.933, precision = 0.964, recall = 0.900, F1 = 0.931, AUC = 0.956), with the lowest RMSE (0.2582), outperforming BO-XGB and all single models. (2) Spatial analysis showed that PSO-XGB identified 92% of known flood points, with the highest density in high-susceptibility zones, and 53% of the study area was classified as low or lowest susceptibility. (3) SHapley Additive exPlanations (SHAP) ranked elevation (0.55), maximum annual daily precipitation (0.51), and population density (0.48) as the top predictors, while drainage network density was a secondary but relevant factor. The proposed framework enhances model interpretability, improves spatial delineation of flood-prone zones, and supports targeted flood risk communication and emergency management.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

A Comparative Study of Urban Pluvial Flood Susceptibility Assessment Based on Multi-Machine Learning Algorithm

  • 期刊论文

  • 发表日期：

    2025年12月29日

查看全文
  • Yaqi Li,
  • Zhengjie Fang,
  • Jun Liu,
  • Zhengsheng Lu,
  • Hong Zhou,
  • Wenhao Yin,
  • Xiaolan Chen

摘要

Rapid urbanization has heightened pluvial flood risks in developing cities, yet most machine learning (ML) studies rely on single-model frameworks with limited comparative evaluation. This study develops a multi-model benchmarking framework for urban pluvial flood susceptibility assessment, integrating nine explanatory variables selected via variance inflation factor (VIF) analysis to reduce multicollinearity. Four ML algorithms—Support Vector Machine (SVM), Random Forest (RF), Gradient Boosting (GB), and eXtreme Gradient Boosting (XGB)—were tested with Bayesian Optimization (BO) and Particle Swarm Optimization (PSO) for hyperparameter tuning, using 200 balanced flood and non-flood samples from Wuxi, China. Key findings are as follows: (1) The PSO-XGB model achieved the highest predictive performance (test accuracy = 0.933, precision = 0.964, recall = 0.900, F1 = 0.931, AUC = 0.956), with the lowest RMSE (0.2582), outperforming BO-XGB and all single models. (2) Spatial analysis showed that PSO-XGB identified 92% of known flood points, with the highest density in high-susceptibility zones, and 53% of the study area was classified as low or lowest susceptibility. (3) SHapley Additive exPlanations (SHAP) ranked elevation (0.55), maximum annual daily precipitation (0.51), and population density (0.48) as the top predictors, while drainage network density was a secondary but relevant factor. The proposed framework enhances model interpretability, improves spatial delineation of flood-prone zones, and supports targeted flood risk communication and emergency management.