The solar radiation prediction is one of the conditions of successful introduction of photovoltaic systems into modern power grids. The nonlinearity and intermittency nature of the solar irradiance necessitated by complexes of meteorological forces and time have a problem with traditional statistical and physical forecasting models. This paper involves a detailed performance assessment of the machine learning models of artificial intelligence based on solar radiation prediction, on the basis of a real meteorological dataset. The use of a systematic methodology entailing data cleaning, time-based feature extraction, feature engineering and feature selection helps to improve data quality and model learning potential. A number of machine learning models, namely K-Nearest Neighbors (KNN), Extra Trees, Random Forest Regressor, and XGBoost are found and tested through standard regression measurements of MAE, MSE, RMSE and R 2 score. The experimental findings reveal that the overall level of the nonlinear and ensemble-based models is far much better than the conventional ones, and the best predictive accuracy (R 2 = 0.946) is observed to be obtained with the KNN model. The results indicate the suitability of the instance-based and ensemble learning method in describing intricate solar radiation patterns, which is likely to provide enhanced forecasting accuracy when planning a renewable energy resource and grid stability.