The problem of supernova photometric identification is still an open issue faced by large photometric surveys. In a previous investigation, we showed how combining Kernel Principal Component Analysis and Nearest Neighbour algorithms enable us to photometrically classify supernovae with a high rate of success. In the present work, we demonstrate that the introduction of Gaussian Process Regression (GPR) in determining each light curve highly improves the efficiency and purity rates. We present detailed comparison with results from the literature, based on the same simulated data set. The method proved to be satisfactorily efficient, providing high purity (⩽ 96%) rates when compared with standard algorithms, without demanding any information on astrophysical properties of the local environment, host galaxy or redshift.