Greenhouse gases time series prediction using support vector regression coupled with decomposition techniques (DWT vs EMD vs VMD)

Predicting greenhouse gas (GHG) emissions allows countries to plan their future GHG reduction activities effectively. In existing machine learning-based studies, GHG emission prediction models have been constructed, ignoring the specific characteristics of collected data, which negatively impacts the prediction performance of these models. This study considered these characteristics by applying data decomposition techniques to preprocess collected data before applying machine learning techniques. To demonstrate the effectiveness of this approach, we investigated the potential and performance of numerous data decomposition techniques, particularly discrete wavelet transform (DWT), empirical mode decomposition (EMD), and variational mode decomposition (VMD), associated with support vector regression (SVR) for predicting GHG time series. The models were built and tested using available data on GHG concentrations from four sites in California, USA. In addition, the accuracy of the models was compared with that of a conventional SVR model. The models developed were then used to predict GHG time series. The results showed that the hybrid model (VMD-SVR) outperformed all other considered models in accuracy and generalizability. The improvement from using this model compared to the conventional SVR model was evident, with average reductions in the root mean square error (RMSE) and mean absolute error (MAE) of 76.33% and 79.29%, respectively. The coefficient of determination (R2) and Nash–Sutcliffe efficiency (NSE) improved by an average of 97.06% and 97.25%, respectively. Consequently, using the model under consideration potentially enables better planning for future related actions.