Machine learning of material properties: Predictive and interpretable multilinear models