fig3
Figure 3. Descriptor sniping and optimal models. (A) The changes in R2 and RMSE for model-A (5-fold-CV), built using RFE nested CatBoost with different numbers of descriptors, were accurately described. Scatter plots of the best model to predict the T5% values versus their experimental values for the train and test sets for (B) model-A and (C) model-D. RMSE: Root mean square error; 5-fold-CV: five-fold cross-validation; RFE: recursive feature elimination.
