The featurization method accurately predicted second virial coefficients with a relative error of less than 1% in the interpolative regime and had a relative error of 2.1% when extrapolating to temperatures outside of the training range for each compound in the dataset.