Magma chemical and physical properties are key to understand magma ocean dynamics, mantle formation and atmosphere generation. For instance, volatile elements solubility in silicate melts influence the magma ocean – atmosphere exchanges, while magma viscosity, density, heat capacity and liquidus will influence the dynamics and crystallization history of a magma ocean. As a result, precise predictive models of magma properties thus are cornerstone to develop our research regarding past and present magma oceans, for instance at the surface of « lava planets ». I will present here our research efforts to develop such models, based on a semi-physical approach that leverages existing experimental data and machine learning. Using a newly implemented model, we will explore how the conditions at the surface of the exoplanet 55 Cancri e may vary depending on surface conditions and magma ocean chemical composition.