The Earth’s outer radiation belt response to geospace disturbances is extremely variable spanning from a few hours to several months. In addition, the numerous physical mechanisms, which control this response, depend on the electron energy, the time-scale and the various types of geospace disturbances. As a consequence, the various models that currently exist are either specialized, orbit-specific data-driven models, or sophisticated physics-based ones. In this paper we present a new approach for radiation belt modelling using Machine Learning methods driven solely by solar wind speed and pressure, Solar flux at 10.7 cm and the $\theta$ angle controlling the Russell-McPherron effect. We show that the model can successfully reproduce and predict the electron fluxes of the outer radiation belt in a broad energy (0.033–4.062 MeV) and L-shell (2.5–5.9) range and, moreover, it can capture the long-term modulation of the semi-annual variation. We also show that the model can generalize well and provide successful predictions, even outside of the spatio-temporal range it has been trained with, using >0.8 MeV electron flux measurements from GOES-15/EPEAD at geostationary orbit.