<![CDATA[Floods are among the most frequent disasters in regions with high rainfall, including Bogor Regency, creating a need for a system capable of predicting rising water levels early enough for preventive measures to be taken. This study applies the Random Forest Regression algorithm to predict the water level (TMA) one hour ahead using historical hydrological data from the Batu Beulah Water Level Monitoring Station, Bogor, comprising water level, rainfall, and discharge derived from a rating curve, and integrates the resulting model into a Next.js-based web application. Unlike locally run machine learning programs accessible only to technical users, the developed system separates the prediction service (FastAPI) from the web interface so that the output can be used by two actors, namely the station administrator and the general public. The dataset consists of 7,817 hourly records collected throughout 2024 with 31 input features, split chronologically into training, validation, and testing sets with a 70:15:15 ratio to avoid data leakage. The model was trained with 300 decision trees and evaluated using Mean Absolute Error (MAE), Root Mean Square Error (RMSE), and the coefficient of determination (R²), and compared against a persistence baseline. Testing results show an MAE of 0.0957 m, an RMSE of 0.2137 m, and an R² of 0.7918, with RMSE and R² outperforming the baseline. Current water level, calculated discharge, and the three-hour rolling mean of water level were the most influential features. Black Box and White Box testing confirmed that the system's core functions, including login, data management, prediction execution, and the display of Safe/Alert/Danger warning status, operate as intended.]]>
Copyrights © 2026
