Sign in Subscribe
Machine learning

Improving Flood Susceptibility Mapping with Swarm Intelligence and Gradient Boosting Algorithms

Strategic Insights

— 1 min read

Research from Sejong University in South Korea has introduced a novel approach to enhance the accuracy of flood susceptibility mapping (FSM) by optimizing the CatBoost algorithm with two swarm-based metaheuristic algorithms: the Zebra optimization algorithm (ZOA) and the Whale optimization algorithm (WOA). This study addresses the critical gap in determining optimal hyperparameters for machine learning models in FSM, which often lead to suboptimal model performance. The researchers conducted FSM using 13 parameters that affect floods and flood occurrence points as inputs, achieving an accuracy of 84.2% for CatBoost, 85% for CatBoost-WOA, and 87.2% for CatBoost-ZOA.

Key Takeaways:

  • A novel approach to increase the accuracy of flood susceptibility mapping (FSM) has been introduced by optimizing the CatBoost algorithm with two swarm-based metaheuristic algorithms: ZOA and WOA.
  • The research aims to enhance and improve the accuracy of FSM in Shushtar County, southwest Iran, using the CatBoost-WOA and CatBoost-ZOA algorithms.
  • The evaluation results of the flood susceptibility maps showed an accuracy of 84.2% for CatBoost, 85% for CatBoost-WOA, and 87.2% for CatBoost-ZOA.
  • The results demonstrated a 3.0% absolute improvement in accuracy with the ZOA-optimized model over the non-optimized CatBoost.
  • Integrating swarm-based methods with machine learning boosting algorithms significantly enhanced FSM accuracy.
  • The research can help managers and decision-makers in flood management and control as a non-structural approach.
  • Seyed Vahid Razavi-Termeh, Abolghasem Sadeghi-Niaraki, Sani I. Abba, Jamil Hussain, and Soo-Mi Choi contributed to the research; Seyed Vahid Razavi-Termeh was the lead researcher.
  • The study was conducted at Sejong University's Department of Computer Science and Engineering and Convergence Engineering for Intelligent Drone, XR Research Center in Seoul, South Korea.

Statistics:

  • Accuracy of flood susceptibility maps: 84.2% for CatBoost, 85% for CatBoost-WOA, and 87.2% for CatBoost-ZOA.
  • Absolute improvement in accuracy with the ZOA-optimized model over non-optimized CatBoost: 3.0%.
  • Number of parameters used for flood susceptibility mapping: 13.
  • Location of the study: Shushtar County, southwest Iran.

Sources:

  • Razavi-Termeh, S. V., et al. "Flood-prone area mapping using a synergistic approach with swarm intelligence and gradient boosting algorithms." Scientific Reports, vol. 15, no. 1, 2025, p. 27924.
  • NewsRx. "Researchers from Sejong University Provide Details of New Studies and Findings in the Area of Machine Learning (Flood-prone area mapping using a synergistic approach with swarm intelligence and gradient boosting algorithms)." Journal of Engineering, August 18, 2025, p. 2588.

Read more