Multivariate methods such as partial least squares (PLS), interval PLS, and other variants are often the default option for prediction of lubricant properties based on FTIR. spectra. However, other advanced analytical methodologies are also available that have not been properly tested and comparatively assessed so far. The present work focuses on the comparison of the predictive ability of four classes of analytical method's: regression with variable selection, penalized regression, latent variable regression, and tree-based ensemble methods. A data set of 62 lubricant samples for different applications was :collected in Portugal. Assessed lubricant properties included kinematic viscosity (at 40 and 100 degrees C), viscosity index, density) total acid number (TAN), saponification number, and percentage of aromatics, naphthenics, and paraffinics. This work showed that there is no overall superior regression method and the choice is dependent on the predicted property. Density, percentage of aromatics, naphthenics, and paraffinics were well predicted (correlation between predicted and observed of 0.97-0.98). Elastic nets was the best method to predict naphthenics and density, but the former property was also well predicted by least absolute shrinkage and selection operator. Interval PLS was the method that provided better prediction of aromatics and paraffirtics. TAN could be reasonably predicted by support vector regression but some clusters were observed. Saponification number and the properties related to viscosity were not satisfactorily predicted with any of the tested methods. Finally, it can be concluded that the adopted methodology is highly relevant in the field of prediction of lubricant oil properties.