Lennard-Jones (LJ) fluid simulations have been widely incorporated during the past decades, establishing Molecular Dynamics (MD) as the most popular alternative to experiments and the calculation of materials properties. The basic idea lies on modelling a system with theoretical structure that can be extrapolated to a real system if appropriate LJ parameters are employed in order to correspond to real fluids. Simulation and experimental results for pure and composite materials are constantly being gathered in electronic databases, most of them freely availalable to the research community, and this opens the pathway for data science to exploit the huge potential offered, towards a data-driven investigation. The incorporation of Machine Learning (ML) techniques in all fields of science and engineering is now a fact. Among various ML methods, Symbolic Regression (SR) has gained a distinct role in interpreting physical phenomena, as it goes through replacing “black-box” predictions with symbolic expressions that can be explained on a physical basis with the extraction of possible underlying laws with no apriori asumptions. Following this direction, this work presents the application of SR on simulation data referring to the calculation of the diffusion coefficient of LJ fluids and the generalization scheme to apply results to real, pure fluids, through symbolic equations of low complexity and error. The proposed equations are compared to existing empirical relations and results suggest that the proposed model performs well both on property prediction but also lies on concrete physical grounds.