Forecasting greenhouse gases emission with hybrid method in Turkey: The common approach of econometrics and machine learning

Every data set has own shape like being in human fingerprint. Because of that in the data analysis process, the data used in analysis has to be known with its detailed structure. Detecting the linear and nonlinear structures, stationary and unstationary structures also to create a process according to this detection has big importance aspect of the correctness of the findings. Otherwise, as a result of using unproper model with the data the general structure of the data would not be detected and thus the findings would be largely inaccurate or lack. With the hybrid method which developed to avoid from this problem to reach the linear and nonlinear structure of the data is available and then the accuracy rate of the model could increase. In this study subjects the total greenhouse gases, CO2, N2O and F gasses related with the carbon emission which the big issue of the recent decades and analysis these gasses with the hybrid method mentioned above. The data covering the period of 1990-2021 years for Türkiye is analysed with data visualization method firstly to detect the structural situation of the data and then with Fourier Augmented Dickey Fuller (FADF) and Augmented Dickey Fuller tests for unit root analysis. The model is firstly created with the support vector regression which one of them among the nonlinear machine learning methods within the framework of autoregressive structure to detect the nonlinear trends and then the autoregresive movement avarage method is employed with using the residuals of the support vector regression for detecting the linear trends of the data. The results obtained from the estimation are evaluated for performance using the criteria of absolute error mean, percentage of absolute error mean, mean square error and square root of the mean square error. According to the lights of the findings, it is understood by the visual and unit root analysis for Türkiye that the greenhouses gasses having the nonlinear and nonstationary structural situations. On the other hand it is seen that to obtain much more correct predicted outputs, to estimate the data using hybrid method can be more proper method contrary to the just using linear or nonlinear methods.