fig4

Figure 4. Feature screening procedure and results of feature selection. (A) Pearson correlation coefficients of the candidate features after eliminating strongly correlated variables; (B) Importance ranking of the top 15 selected features, where the blue bars indicate features removed during the exhaustive feature selection process; (C) Exhaustive feature selection results showing the variation of model performance (10-fold cross-validated R²) with the number of selected features; (D) Comparison between predicted and experimentally measured tensile strength for the optimal model. R²: The coefficient of determination; RMSE: root mean square error; MAE: mean absolute error.