<![CDATA[Rainfall data in Samarinda City exhibit seasonal patterns that play a crucial role in increasing flood risk during certain periods. To enhance the effectiveness of the early warning system, this study developed a hybrid SARIMA-RBFNN model. The Seasonal Autoregressive Integrated Moving Average (SARIMA) model was applied to capture linear seasonal patterns, while the Radial Basis Function Neural Network (RBFNN) was employed to model nonlinear residuals from SARIMA. Model performance was assessed using the Symmetric Mean Absolute Percentage Error (SMAPE) and Root Mean Squared Error Prediction (RMSEP). Compared to the single SARIMA model (SMAPE = 34.699%, RMSEP = 82.255), the hybrid SARIMA–RBFNN achieved lower in-sample errors (SMAPE = 34.175%, RMSEP = 78.577) and demonstrated more stable performance for out-of-sample data. This indicates that the hybrid model provides a more balanced and reliable prediction by capturing nonlinear rainfall fluctuations that SARIMA alone could not model effectively. Forecasts for 2024 revealed a consistent seasonal trend, peaking mid-year. These findings indicate that the hybrid model can improve the reliability of the flood early warning system in Samarinda by providing more accurate rainfall predictions.]]>
Copyrights © 2025
