Mechentel et al. (2026) Data-driven flood susceptibility assessment using hybrid machine learning and optimization techniques: case of the Sedrata Watershed, NE Algeria

Identification

• Journal: Scientific Reports
• Year: 2026
• Date: 2026-04-04
• Authors: Elhadi Mechentel, Sabri Dairi, Abdelouahab Lefkir, Saeid Eslamian, Habib Abida, Yassine Djebbar
• DOI: 10.1038/s41598-026-43262-9

Research Groups

• Department of Hydraulics, Badji Mokhtar University, Annaba, Algeria
• Laboratory of Research Infra-Res, Mohamed Cherif Messaadia, Souk Ahras, Algeria
• École Nationale Supérieure des Travaux Publics (ENSTP), Kouba, Algiers, Algeria
• Department of Water Science and Engineering, College of Agriculture, Isfahan University of Technology, Isfahan, Iran
• GEOMODELE Laboratory, Faculty of Sciences, University of Sfax, Sfax, Tunisia

Short Summary

This study developed a hybrid flood susceptibility modeling framework combining Random Forest Regressor with metaheuristic optimization algorithms (GOA, SSA, ACO) for the Sedrata Watershed, Algeria, demonstrating significantly enhanced predictive performance compared to the standalone RFR model.

Objective

• To develop and evaluate a hybrid flood susceptibility modeling framework using Random Forest Regressor coupled with metaheuristic optimization algorithms (Grasshopper Optimization Algorithm, Salp Swarm Algorithm, and Ant Colony Optimization) to provide reliable assessment tools for informed flood risk management.

Study Configuration

• Spatial Scale: Sedrata Watershed, Northeastern Algeria.
• Temporal Scale: Flood locations were detected from Sentinel-1 imagery; no specific study period or temporal range for the flood events is explicitly stated.

Methodology and Data

• Models used: Random Forest Regressor (RFR), RFR-Grasshopper Optimization Algorithm (RFR-GOA), RFR-Salp Swarm Algorithm (RFR-SSA), RFR-Ant Colony Optimization (RFR-ACO). Weight of Evidence and Geographically Weighted Regression were used to investigate conditioning factor influence.
• Data sources: Satellite data (Sentinel-1 imagery for 317 flood locations), and twelve physiographic and environmental conditioning factors including slope, rainfall, land use, drainage density, curvature, convexity, and aspect.

Main Results

• The hybrid models significantly enhanced predictive performance compared to the standalone RFR model.
• Area Under Curve (AUC) values for the hybrid models were: RFR-ACO (0.928), RFR-SSA (0.925), and RFR-GOA (0.920).
• The baseline standalone RFR model achieved an AUC of 0.904.
• Spatial mapping revealed that approximately 26% of the study area exhibits high to very high flood susceptibility.

Contributions

• Proposes a novel hybrid flood susceptibility modeling framework by integrating the Random Forest Regressor with metaheuristic optimization algorithms (GOA, SSA, ACO).
• Demonstrates a significant improvement in the predictive performance of flood susceptibility maps through the use of these hybrid models.
• Offers an interpretable decision-support tool for sustainable flood risk management and land use planning, specifically for the Sedrata region.
• Confirms the strong potential of combining Machine Learning and metaheuristic optimization techniques for enhancing flood susceptibility mapping.

Funding

• This research received no specific grant from any funding agency in the public, commercial, or not-for-profit sectors.

Citation

@article{Mechentel2026Datadriven,
  author = {Mechentel, Elhadi and Dairi, Sabri and Lefkir, Abdelouahab and Eslamian, Saeid and Abida, Habib and Djebbar, Yassine},
  title = {Data-driven flood susceptibility assessment using hybrid machine learning and optimization techniques: case of the Sedrata Watershed, NE Algeria},
  journal = {Scientific Reports},
  year = {2026},
  doi = {10.1038/s41598-026-43262-9},
  url = {https://doi.org/10.1038/s41598-026-43262-9}
}