Enhanced daily streamflow forecasting in Northeastern Algeria: integrating hybrid machine learning with advanced wavelet transformation techniques

The primary goal of this study is to devise robust models for analyzing daily streamflow time series across three distinct watersheds in northeastern Algeria, employing artificial intelligence techniques. The approach integrates four predictive models: Multi-Layer Perceptron Neural Network (MLPNN), Extreme Learning Machine (ELM), Random Forest Regression (RFR), and M5 Tree Model (M5Tree). A novel modeling technique introduced herein leverages the Maximum Overlap Discrete Wavelet Transform (MODWT) for preprocessing the input variables. This technique decomposes the inputs into multiple sub-signals, which then serve as new inputs for the machine learning models. The enhanced models, particularly MODWT-MLPNN and MODWT-M5Tree, demonstrated superior numerical performance, achieving correlation coefficients (R) of 0.994 and 0.989 and Nash-Sutcliffe Efficiency (NSE) scores of 0.985 and 0.977, respectively. These results underscore the effectiveness of the decomposition method in surpassing the accuracy of standalone models.