Flood susceptibility modeling based on new hybrid intelligence model: Optimization of XGboost model using GA metaheuristic algorithm

Flood is the most common natural hazard that causing unprecedented loss of life and property in the world. In recent years, flood damage has increased due to human intervention and land use and climate changes. The purpose of this study is to predict flood susceptibility in Tafresh watershed in Markazi Province, Iran based on K-nearest neighbor (KNN), Extremely Gradient Boosting (XGB) machine learning models and evaluate the performance of hybrid genetic algorithm (GA) optimization method and XGB model. For this purpose, 14 independent variables affecting flood susceptibility were prepared, also, 227 flood locations were identified as independent variables based on available information and field survey. In order to evaluate the efficiency of the models, receiver operating characteristic (ROC) parameters were used. Evaluating the efficiency of the models based on the AUCs of testing dataset showed the higher efficiency of the GA-XGB hybrid model in modeling flood susceptibility in the Tafresh watershed is compared to KNN and XGB models, and the AUC in KNN, XGB, and GA-XGB models are 0.82, 0.85, and 0.87, respectively. So using the genetic algorithm as an optimizer for determining the best parameters in the XGB model increases efficiency in this model. The results of determining the relative importance of independent variables in flood susceptibility modeling showed that the independent variables considered in each model have different effects. Distance from road and distance from river in all three models had significant importance in modeling flood hazard in the Tafresh watershed. The optimization method in this study can be used as a powerful method in other spatial modeling studies. The results of this study also support management programs to reduce flood risks in the study area. (C)& nbsp;2022 COSPAR. Published by Elsevier B.V. All rights reserved.

Automated summary

This study predicts flood susceptibility in the Tafresh watershed in Iran using machine learning models and optimizes the model parameters with a genetic algorithm. The results show that the optimized GA-XGB hybrid model has higher efficiency in flood susceptibility modeling and can support flood risk management programs.