The increased interest in the sources of renewable electricity has drawn attention to the rapidly developing solar energy sector owing to its high cost-benefit ratio. To accurately calculate the potential electricity output of photovoltaic (PV) panels, the global horizontal irradiance (GHI) and diffuse horizontal irradiance (DHI) must be known. The worldwide presence of satellite imagery provides an efficient way to estimate current and historical GHI instead of using the high time- and cost-consuming in-situ measurement. The most of past studies utilized more satellite bands as the inputs for machine learning algorithms to estimate GHI. The further estimation of DHI was usually based on weather-related parameters. Thus, there is a research gap in using only one satellite band for both GHI and DHI estimations. Therefore, this study used machine learning algorithms to estimate GHI and DHI, with inputs delivered from the Heliosat model based on band 3 of Himawari-8 satellite imageries. The results were compared with the original and site-adapted Heliosat models and seven DHI separation models. The results indicated that the machine learning models were capable of performing with the same accuracy as the Heliosat models. However, their performance was better while estimating DHI, in which case they outperformed even the best separation model. Higher accuracy and precision were observed in those models where the additional solar zenith at time t+1h was used together with other input features. This highlighted the possibility of using only one satellite band together with the calculated solar position variables as the input. Overall, this research has established a new method for estimating GHI and DHI with high confidence based on satellite imageries and the Heliosat model through the application of machine learning techniques.