Combining Machine Learning Algorithms with Empirical Mode Decomposition and Discrete Wavelet Transform for Monthly Peak Discharge Prediction

Abstract

Accurate and reliable peak discharge prediction is of great importance in water resource management and flood control studies. The aim of this study was to predict monthly peak discharge by combining various signal decomposition processes and machine learning models. For this purpose, monthly peak discharge data were decomposed into subsignals utilizing Daubechies 3, Coiflet 5, discrete Meyer main wavelets, and empirical mode decomposition methods, which are the ones most commonly used in hydrological studies. The separated signals were subjected to correlation analysis, and the sum of the highly correlated signals was presented as input to the machine learning models. Support vector machines, regression trees, ensemble trees, and adaptive neuro-fuzzy inference system models were used for peak discharge forecasting. The performance of the established models was evaluated with the help of statistical indicators such as mean absolute error, root mean squared error, and determination coefficient, and graphically through Taylor diagrams. At the end of the study, the most effective results were obtained with the hybrid model established by the Daubechies 3 wavelet-4 decomposition levels and combined with the coarse Gaussian support vector machine.