Classical transferable force fields allow predicting physical properties and microscopic processes of pure substances and mixtures. The development of force fields, especially the parametrization of the ‘non-bonded’ interactions, requires multidimensional optimization with computationally expensive (and noisy) objective function evaluations through Monte Carlo (MC) and molecular dynamics (MD) simulations. This is demanding because force-field parameters are highly correlated. We use physically-based surrogate models to drastically accelerate the optimization procedure and specifically propose new reduced order models for transport properties based on entropy scaling. Transport properties, such as viscosity, thermal conductivity, and self-diffusion coefficients, can then be considered in the optimization of force fields simultaneously with static properties, which was not possible before. Machine-learned models for transport properties will be developed with an increasing data-base of simulated force field parameters and substances.