Performance Comparison of Random Forest, XGBoost, and SVM for Flood Risk Prediction Using BNPB GIS Data

This study compares the performance of three machine learning algorithms—Random Forest, XGBoost, and Support Vector Machine (SVM)—for predicting flood risk using spatial and non-spatial data from BNPB GIS. The analysis focuses on disaster records from January 3 to 15, 2026, with district-city as the spatial unit of observation. Following data cleaning, exploratory analysis, and feature preparation, the models were evaluated using ROC-AUC, PR-AUC, F1-Score, Precision, Recall, and Accuracy. XGBoost demonstrated the highest ROC-AUC (0.675), indicating strong overall performance in distinguishing flood from non-flood events. Random Forest achieved the highest Recall (0.947), showing superior sensitivity in detecting flood events, while SVM exhibited fluctuating performance with a lower ROC-AUC (0.496). Visualizations of model behavior and spatial flood patterns were provided to support model interpretability. The study’s results suggest that ensemble models, particularly XGBoost and Random Forest, can significantly enhance flood risk prediction, improving the accuracy and sensitivity of early warning systems. These findings contribute to the development of more effective data-driven flood mitigation strategies in Indonesia, enabling better disaster preparedness and response.