Answers are grounded in this paper and cite their sources.
This research compares hybrid CNN and XGBoost models for flood prediction, indicating effective mapping techniques in urban areas.
• This research applied XGBoost, CNN, and metaheuristic algorithms for flood prediction. • Multiple statistical metrics (AUC, RMSE, MAE) were used for model evaluation. • XGB–HHO achieved the highest AUC (98.3%) and lowest RMSE (0.216). • Hybrid models demonstrated exceptional precision in mapping flood susceptibility. Quantitative modeling for flash flood susceptibility mapping (FFSM) is a practical, realistic approach that can substantially mitigate the adverse effects of this natural disaster. Accordingly, the objective of this study is to combine and compare the strengths of two potent methods, i.e., CNN and XGBoost, with three metaheuristic algorithms, i.e. ABC, HHO, and HOA, to determine the best combination. Historical flood data, social media data, and relevant reports were employed to identify 157 potential flood-prone areas ( i.e., dependent variable). Independent variables for flood occurrence included slope, elevation, distance to river, geology, river density, rainfall, land cover, SPI, and TWI. Furthermore, to evaluate the predictive performance of the different models, the ROC curve and statistical metrics such as sensitivity, specificity, accuracy, RMSE, and MAE were calculated. The validation outcomes showed that all models performed exceptionally well in flood prediction with only minimal differences in their performance. In particular, XGB–HHO outperformed the other models, exhibiting the lowest RMSE (0.216) and highest AUC (98.3%). The XGB–ABC model yielded the lowest MAE (0.103). The corresponding AUC values for the XGB-ABC, XGB-HOA, CNN-HHO, CNN-ABC, and CNN-HOA models were 97.8%, 97.7%, 95.9%, 95.8%, and 95.8%, respectively. The proposed hybrid models can serve as powerful and precise methods in other flood susceptibility mapping studies.
No takes yet. Share an insight, caveat, or question.
Neshat et al. (2026) studied this question.
Synapse has enriched 10 related papers. Consider them for comparative context: