Flood Prediction Model Using the Random Forest Algorithm in Padangsidimpuan City

Abstract

Flooding is a complex environmental phenomenon influenced by meteorological, temporal, and urban conditions. This study aims to develop a flood risk prediction model in Padangsidimpuan City using the Random Forest algorithm by integrating meteorological and temporal variables. The data set consists of historical observations from Padangsidimpuan City, including rainfall, temperature, humidity, wind direction, and flood occurrence status. The data were processed and divided into training and testing sets to evaluate model performance. The results indicate that the Random Forest model achieves strong classification performance, with an accuracy of 0.98 and specificity of 0.99, demonstrating a high capability in correctly identifying non-flood conditions. The model also shows good discrimination ability, with a ROC-AUC value of approximately 0.87. However, the recall value of 0.50 suggests that only half of the actual flood events are correctly detected, primarily due to class imbalance between flood and non-flood data. Feature importance analysis reveals that short-term meteorological variables, particularly rainfall and temperature, along with temporal patterns, are the most influential factors in flood prediction. In addition, spatial interpretation shows that flood-prone areas in Padangsidimpuan City are concentrated in densely populated urban zones with limited drainage capacity, highlighting the influence of urban environmental conditions on flood risk.Overall, the Random Forest model provides a strong foundation for flood risk prediction in Padangsidimpuan City. However, further improvements in data balancing and model optimization are required to enhance sensitivity and support a more reliable flood early warning system.