Information on solar radiation variability over specific areas is a key element for the design and optimization of PV power generating systems. Variation of the energy output mainly by irradiance/temperature changes of such devices originates from fluctuations of meteorological conditions, seasons and time of the day. Solar radiation is one of the mainly driver that influence the perormance of solar cell/module. As a result, here in, twelve ML models are developed using ground-based hourly observations from 2020 to 2022. For this purpose, sixteen Ethiopian synoptic weather station solar radiation data were used to train the ML models using meteorological, aerosol-related optical obtained from MERRA-2 reanalysis model, sky condition from NASA Power data archive, and other from National Meteorological Institute of Ethiopia (NMI). The data set is partitioned such that 75% were used for model training and 25% used for model testing. A stacked model was then developed to predict solar radiation utilizing the most efficient of these techniques. The stacked model and XGB show superiority in reproducing observed solar radiation data with highest correlation coefficient (R2=0.96; 0.95), lowest root mean squared error (RMSE=11.8W/m2; 9.94 W/m2), and lowest mean average error (MAE=9.7W/m2; 8.4W/m2) for daily and monthly time scales.