Solar rays on earth surface are essential for water resources, environmental and agricultural issues. They severely impact plant growth, global temperatures, mean sea levels, and climatic extremes. The most critical challenge in renewable energy applications is accurate solar radiation (Rs) forecasting. It has become increasingly important in directing solar energy conversion systems, selecting appropriate regions, and even making decision in future investment strategies. The most significant disadvantage of this energy is that it is nonlinear and works only when the load requires it. Since measurement tools are expensive, difficult to install, and not available in all regions of the world, as well as other concerns like equipment failure or water and dirt building up in the sensors, which might cause data to be missing or out of the expected range. Rs is seldom recorded at weather stations. As a result, several methodologies, including empirical, statistical, satellite, time series-based models, and others, have been employed to estimate it using other environmental parameters. Many approaches and preprocessing models based on machine learning (ML) are validated for Rs forecasting and complexity reduction to reduce these concerns. In this paper, we present a model that shows the impact of feature selection (FS) on time series for accurate Rs forecasting. The proposed model use recursive feature elimination (RFE) with random forest (RF), decision tree (DT), logistic regression (LR), classification and regression tree (CART), person (Pr) and gradient boosting models (GBM). The obtained results show that the models based on CART, LR, and GBM provide high accuracies of 0.894, 0.884, and 0.882, respectively. These three accurate methods have same standard deviation (std) of 0.033. In terms of the negative mean absolute error (nMAE) (std), the methods provide 0.892 (0.029), 0.885 (0.034), and 0.885 (0.035) respectively. Compared with other models, the RFE model provides significant impacts features and good performances.