Use of machine learning models to correct systematic errors in DFT-computed enthalpies of formation by learning error patterns from electronic structure features. Can reduce PBE errors from MAE ~195 meV/atom to ~80 meV/atom.