<![CDATA[Rainfall prediction with very high temporal resolution, such as ten-minute intervals, is of great urgency in urban flood mitigation, especially in densely populated areas such as South Tangerang. Most previous studies still focus on daily or monthly rainfall prediction, so this study attempts to fill this gap by comparing the performance of three machine learning algorithms: Random Forest (RF), Support Vector Machine (SVM), and K-Nearest Neighbors (KNN) in predicting ten-minute rainfall. A dataset of 52,703 entries was obtained from an automated weather station (AWS) with predictor variables of temperature, humidity, and air pressure. The analysis process was carried out through preprocessing stages (missing value imputation, Min–Max normalization), data splitting (80:20), and 10-fold cross-validation using Orange Data Mining software. Model performance evaluation used Root Mean Squared Error (RMSE), Mean Absolute Error (MAE), and coefficient of determination (R²). The results showed that Random Forest performed best with an RMSE of 0.671 and MAE of 0.090, followed by KNN with an RMSE of 0.683 and MAE of 0.083, while SVM performed significantly lower (RMSE of 2.553 and MAE of 2.511). However, the negative R² values for all models indicate limitations in explaining rainfall data variability, which is likely influenced by skewed data distribution (zero-inflated) and AWS sensor noise. These findings indicate that Random Forest is relatively more reliable than other models, but its accuracy is still not optimal for operational applications. Future research is recommended to adopt a hybrid or deep learning approach (e.g., LSTM), add other meteorological variables such as wind speed and solar radiation, and integrate the models into IoT-based early warning systems to support urban flood mitigation.]]>
