Machine learning-based mass density model for hard magnetic 14:2:1 phases using chemical composition-based features

•Mass density of 14:2:1 permanent magnetic phases, which are not readily available in the literature, have been predicted using a machine learning approach.•For such 14:2:1 single phases, conversion of measured magnetic moments (in µB/f.u. or emu) into the magnetic saturation polarization (in Tesla), necessitates mass density, which frequently does not get reported.•We achieved accurate mass density predictions (mean absolute error of 0.5%), using readily available compositional data from the literature. For this purpose, chemical composition ‘C’ and atomic mass ‘AM’ features suffice.•Lattice parameters are not necessary for achieving accurate predictions.