fig12

Figure 12. R² value variation graph of models trained on virtual sample sets generated by multiple algorithms with varying quantities. The evaluated models are (A) the GBR, which performs the best; and (B) the ANN, which shows the highest performance improvement. NNI-SMOTE: Nearest Neighbor Interpolation-Synthetic Minority Oversampling Technique; GMM-VSG: a Virtual Sample Generation algorithm based on Gaussian Mixture Models; VAE: Variational Autoencoder; VSG: Virtual Sample Generation; OK: Ordinary Kriging; GBR: Gradient Boosting Regression; ANN: Artificial Neural Networks.