Machine learning training pipeline.

The three extreme gradient boosting-based models were built through parameter optimization using the training dataset and validation with unforeseen observations, stored in the testing dataset. The best combination of parameters was extracted through 10-fold cross-validation. ‘cs’: centered and scaled dataset. ‘cs-pc’: centered and scaled training dataset with extraction of principal components. ‘cs-h_cor_ou’: centered and scaled training dataset leaving highly correlated variables out.