Prediction of renewable energy consumption rates of OECD countries with machine learning techniques

The issue of energy remains one of the biggest problems given the growing population, production and technological development. The ongoing shift towards renewable sources is attributed to the negative impacts of non-renewable sources used in energy production, such as their limited availability, vulnerability to control by certain authorities, their contribution to climate change and negative impacts on human health . Although the transition between energy sources is no longer as slow as before, a complete transition has not yet been achieved. Despite the challenges of fully transitioning to renewable energy for economic, natural and technological reasons, the efforts in this regard are remarkable. Estimating the share of renewable energy consumption in total energy consumption is crucial both for planning future resources and for effectively managing current resources. In this thesis study, it is aimed to estimate the share of renewable energy consumption in the total energy consumption of OECD countries with machine learning methods, to determine the important variables for this estimation and to determine the optimum number of variables. The study uses machine learning methods, artificial neural networks (YSA), support vector machines (SVM), eXtreme gradient boosting (XGBoost) and multivariate adaptive regression splines (MARS). It was aimed to determine the most successful algorithm with LOPCOW, PROMETHEE II, MAIRCA and VIKOR MCDM methods by using RRMSE, R2, VK, MSE, and MAPE statistical performance criteria of the prediction values obtained from machine learning methods. As a result, it was understood that the XGBoost algorithm made the most successful prediction among the algorithms used. In addition, when examining successful models, it becomes clear that in predicting the share of renewable energy consumption in total energy consumption, regulatory quality, value added of services, military expenditures, merchandise trade, carbondioxide emissions, energy consumption and land area all play a role will variables contribute to the models. Additionally, it was determined that the variables used in all models and important for predictions were carbondioxide emissions and country surface area. The study presents examples from previous research that have used these variables, discusses possible measures to increase renewable energy consumption, and provides suggestions for future research areas based on obtained findings and the literature review.