Fig. 6: The effect of different elemental feature dimensions on the prediction performance of six properties on the validation set.

a shows the R2 of models without data augmentation after PCA downscaling. b shows the cumulative variance percents of elemental features after PCA downscaling. c, d show the impact of the target feature dimensions generated by the GAN model after PCA downscaling on the prediction performance; e, f show the impact of the elemental feature dimensions extracted by the ECNet model on the results.