Data Mining Approach for River Flood Hazard Time-Series: Using a Combination of Triple Exponential Smoothing and Neural Networks, in Demak

Conventional flood hazard maps depict a static perspective on flood risk and are helpful in flood forecasting. Then, time-series flood forecasting research was developed and widely conducted. Still, it has limitations because the univariate method used does not consider the influence of other variables, unlike regression. Also, the determination of flood hazard weights, which is usually conducted through empirical studies using AHP, has a good level of accuracy but has the drawback of expert subjectivity. A new approach is proposed for flood hazard forecasting, combining the Triple ES method with NN and determining weights in a data-driven manner using NN. A combination of time series and regression methods (significantly non-linear), namely Triple ES with NN, results in good accuracy with an MSE value of 3.03, MAE 1.20, RMSE 1.74, R² 0.54, and a MAPE of 32.93%. The evaluation results for flood hazard weight determination with an MSE of 0.0111 and a MAPE of 7.81% show promising results, and the weights can be used in Hazard Flood GIS. Visualization in the form of a GIS Hazard Map can be done after all related raster data have been combined. The computational outcomes, particularly the MSE and MAPE values, demonstrate the effectiveness of the proposed approach, providing a clear understanding of the model's performance.