Scholarlink

<p>One of the most devastating natural calamities, flooding results in significant financial damage as well as casualties. The creation of flood susceptibility maps is critical for effectively identifying high-risk locations prone to flooding, which plays an important role in urban flood management. In order to map flood-prone areas, this study uses three machine learning algorithms—random forest (RF), extreme gradient boosting (XGBoost), and support vector machine (SVM)—in Chengdu City. These algorithms are combined with gray wolf optimizer (GWO), particle swarm optimization (PSO), whale optimization algorithm (WOA), and dung beetle optimizer (DBO) to create hybrid models. The accuracy, recall, precision, area under the receiver operating characteristic curve (AUC), F1 score, and root mean square error (RMSE) of the models were used to assess their predictive performance. DBO-RF (AUC = 0.877) had slightly better prediction capability than the other models under comparison. The AUC values for the single model ranged from 0.827 to 0.850, while the remaining hybrid models ranged from 0.856 to 0.876. The RF-GWO (4125&#xa0;s) hybrid model had the longest runtime of all the hybrid models, while the SVM-WOA (691&#xa0;s) hybrid model showed the smallest duration. The flood susceptibility modeling findings show that combining meta-heuristic techniques and machine learning models considerably improves forecast performance. The predicted accuracy of the hybrid models is consistently higher than that of the individual machine learning models. Particularly, in terms of prediction metrics like accuracy and AUC as well as computational efficiency, as shown by shorter runtime, the hybrid model optimized with DBO and WOA algorithms performs better than the model optimized with GWO and PSO algorithms. The remarkable global search capabilities of DBO and WOA algorithms in hyperparameter optimization tasks are responsible for this advantage. In summary, this study provides a clear and effective solution to this problem by validating the effectiveness of the hybrid framework of machine learning and meta-heuristic algorithms in flood vulnerability modeling.</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 hybrid approach combining meta-heuristic algorithm and machine learning for flood susceptibility mapping

  • 期刊论文

  • 发表日期：

    2026年7月14日

查看全文
  • Lina Mo,
  • Xiaogang Qi,
  • Yize Cai,
  • Cunzhe Lu,
  • Jiahui Li,
  • Lifang Liu

摘要

One of the most devastating natural calamities, flooding results in significant financial damage as well as casualties. The creation of flood susceptibility maps is critical for effectively identifying high-risk locations prone to flooding, which plays an important role in urban flood management. In order to map flood-prone areas, this study uses three machine learning algorithms—random forest (RF), extreme gradient boosting (XGBoost), and support vector machine (SVM)—in Chengdu City. These algorithms are combined with gray wolf optimizer (GWO), particle swarm optimization (PSO), whale optimization algorithm (WOA), and dung beetle optimizer (DBO) to create hybrid models. The accuracy, recall, precision, area under the receiver operating characteristic curve (AUC), F1 score, and root mean square error (RMSE) of the models were used to assess their predictive performance. DBO-RF (AUC = 0.877) had slightly better prediction capability than the other models under comparison. The AUC values for the single model ranged from 0.827 to 0.850, while the remaining hybrid models ranged from 0.856 to 0.876. The RF-GWO (4125 s) hybrid model had the longest runtime of all the hybrid models, while the SVM-WOA (691 s) hybrid model showed the smallest duration. The flood susceptibility modeling findings show that combining meta-heuristic techniques and machine learning models considerably improves forecast performance. The predicted accuracy of the hybrid models is consistently higher than that of the individual machine learning models. Particularly, in terms of prediction metrics like accuracy and AUC as well as computational efficiency, as shown by shorter runtime, the hybrid model optimized with DBO and WOA algorithms performs better than the model optimized with GWO and PSO algorithms. The remarkable global search capabilities of DBO and WOA algorithms in hyperparameter optimization tasks are responsible for this advantage. In summary, this study provides a clear and effective solution to this problem by validating the effectiveness of the hybrid framework of machine learning and meta-heuristic algorithms in flood vulnerability modeling.