Performance of predictive models fitted to the training set (60%) and evaluated on the validation set (20%)
| Model type | No. of features | AUC (%) | Average precision (%) | Brier score | PPV (%) | Sensitivity (%) | Precision (%) | Youden Index (%) | F1 score | NPV (%) | Specificity (%) |
| XGBoost model | 479 | 79.4 | 75.6 | 0.174 | 73.5 | 61.7 | 68.8 | 43.0 | 0.651 | 76.1 | 81.4 |
| Reg logistic model | 458 | 76.5 | 71.5 | 0.187 | 71.5 | 60.2 | 65.7 | 39.2 | 0.628 | 74.8 | 79.0 |
| Logistic model | 481 | 76.4 | 71.5 | 0.187 | 71.4 | 60.2 | 65.5 | 39.0 | 0.627 | 74.8 | 78.8 |
| Random forest model | 478 | 76.1 | 72.3 | 0.198 | 71.9 | 57.5 | 67.4 | 39.0 | 0.621 | 74.2 | 81.4 |
| Lasso model | 58 | 75.3 | 70.9 | 0.194 | 72.0 | 54.0 | 69.3 | 38.0 | 0.607 | 73.3 | 84.1 |
Reg logistic model, logistic regression model; higher AUC, better distinction between patients with and without obesity; lower Brier score, greater accuracy; higher recall, better maximization of the number of true positives; higher precision, better minimization of false positives; Youden Index of >50%, higher F1 score, better performance of model; higher specificity, better identification of negative results.
AUC, area under the curve; Lasso, Least Absolute Shrinkage and Selection Operator; NPV, negative predictive value; PPV, positive predictive value; XGBoost, extreme gradient boosting.