Comparison of The Performance of Machine Learning Methods for Solar Energy Power Prediction

In parallel with population growth, the demand for electrical energy continues to rise. As traditional energy sources have proved insufficient to meet this demand, attention has shifted to environmentally friendly renewable energy sources, among which solar energy stands out. Electricity generation from solar energy is achieved via photovoltaic (PV) systems. Accurately predicting generated power using machine learning (ML) methods with low error supports effective and efficient electricity planning. This study aims to predict solar energy power using ML methods—Gradient Boosting Regressor (GBR), Decision Tree Regressor (DTR), Random Forest Regressor (RFR), and AdaBoost Regressor (AR). For this purpose, the “Solar energy power generation” dataset from Kaggle was utilized. The dataset includes meteorological variables such as temperature, humidity, and radiation, consisting of 21 variables and 4,212 measurements. The Relief feature selection method was used to identify the most informative independent variables for predicting solar energy power. Seventy percent of the data was used for training and the remainder for testing. Model performance was examined with respect to the number of variables. To compare methods, we employed MSE, RMSE, MAE, R2, and training time, and applied cross-validation to enhance model performance. The results indicate that AR achieved superior predictive performance, yielding lower errors (MSE = 0.111, RMSE = 0.333, MAE = 0.145) than the other methods. In contrast, RFR was found to be better in terms of both speed (training time = 2.896) and explained variance (R2=0.926). Additionally, except for DTR, all evaluated techniques exhibited improved performance as the number of variables increased.