A comprehensive study on the applicability of kernel ridge regression for predicting atomization energies

Motivated by the previous study by Rupp et al. [Phys. Rev. Lett, 108, 058301], we implement an improved Kernel Ridge Regression (KRR) model trained on the QM7 dataset and provide a comprehensive evaluation of its applicability. The model is further applied to predict the stability difference between the cis- and trans-isomers of 1,2-dichloroethylene. The accuracy of our model is comparable to that of density functional theory (DFT) calculations, achieving a mean absolute error (MAE) of approximately 4.4 kcal/mol on QM7 dataset. Our findings suggest that, although the KRR model is well established, its performance and transferability can be substantially enhanced through careful feature engineering.