Hybrid least squares support vector machine for water level forecasting

Previous studies have highlighted the significant role of historical water level data in flood forecasting. In this study, we compare two standalone models, Support Vector Machine (SVM) and Least Squares Support Vector Machine (LSSVM), with hybrid models that integrate Ensemble Empirical Mode Decomposition (EEMD) with SVM and LSSVM, aiming to develop a more effective forecasting approach for hydrological data. Particle Swarm Optimization (PSO) is incorporated into these hybrid models to optimize the parameters of SVM and LSSVM, resulting in four models: SVM-PSO, LSSVM-PSO, EEMD-SVM-PSO, and EEMD-LSSVM-PSO. This study focuses on forecasting water levels in Sungai Gombak, Malaysia. The performance of the proposed models is evaluated and compared using several metrics, including RMSE, MSE, MAPE, and the squared correlation coefficient. Results indicate that the EEMD-LSSVM-PSO model outperforms the other models, demonstrating the highest forecasting accuracy for Sungai Gombak, Malaysia, with the lowest RMSE, MSE, and MAPE values and the squared correlation coefficient value close to 1 for the testing data.