Disaster

[D4-03] Assessing Urban Flood Susceptibility Maps Based on Machine Learning Models in Seoul, South Korea

*Julieber Toralde Bersabe¹, Byong-Woon Jun¹ (1. Kyungpook National University)

Keywords:Flood Susceptibility, Machine Learning, Random Forest, Logistic Regression, Support Vector Machine

This study assessed flood susceptibility maps using logistic regression (LR), random forest (RF), and support vector machines (SVM) models in Seoul metropolitan city, South Korea. We constructed a flood inventory map consisting of 1,000 flood points and 1,000 non-flood points. These datasets were randomly divided into training and testing datasets in a 70:30 ratio. In addition, we selected 16 variables as flood conditioning factors considering previous literature and data availability. Various evaluation metrics such as accuracy, kappa, and AUC were computed to comprehensively evaluate the models' accuracy and predictive capabilities. The resulting flood susceptibility map was classified into five categories, ranging from very low to very high susceptibility. By comparing the evaluation metrics for each model, we concluded that the RF model outperformed the LR and SVM models, exhibiting an accuracy of 0.837 and an AUC of 0.902. Furthermore, our findings highlighted that the most significant flood conditioning variables were sewer pipe density, distance to storm drain, elevation, rainfall, and terrain ruggedness index.