<![CDATA[This study developed a water level prediction model using the integration of Artificial Neural Network (ANN) and Long Short-Term Memory (LSTM). ANN maps nonlinear relationships between hydrological parameters, while LSTM captures temporal dependence patterns in time series data. The comparison results of four models—Linear Regression, ANN, LSTM, and ANN–LSTM hybrid—showed that the neural network-based model provided significantly better prediction performance than the linear model. The Linear Regression model produced the largest error (MSE 0.0175; RMSE 0.128; MAE 0.103; R² 0.785), followed by ANN with a significant increase in accuracy (MSE 0.0114; RMSE 0.101; MAE 0.0549; R² 0.874). LSTM provided better results (MSE 0.0048; RMSE 0.067; MAE 0.0472; R² 0.902), but the best model was the hybrid ANN–LSTM with the lowest error value (MSE 0.00417; RMSE 0.063; MAE 0.0388) and the highest R² (0.937). This combination is able to capture nonlinear patterns and temporal dynamics more optimally, resulting in stable and accurate predictions. In addition, this study shows that the Duflow hydrodynamic model has the potential to be developed as a mitigation simulation tool for water level management, such as testing extreme rainfall scenarios, spatial changes, flood control infrastructure operations, channel optimization, and simulating the impacts of climate change.]]>
Copyrights © 2026
