Table 5.
Discriminative power and calibration.
| Methods | AUROC overall (n = 156,750) | Brier score overall (n = 156,750) | AUROC obesity (n = 16,818) | AUROC diabetes (n = 32,991) |
|---|---|---|---|---|
| Logistic Regression | 0.629 ± 0.01 (0.604-0.654) | 0.006 ± 0 (0.0063-0.0063) | 0.619 ± 0.03 (0.602-0.636) | 0.583 ± 0.07 (0.526-0.640) |
| XGBoost | 0.601 ± 0.03 (0.578-0.624) | 0.006 ± 0.0002 (0.0063-0.0066) | 0.567 ± 0.03 (0.540-0.594) | 0.590 ± 0.05 (0.549-0.630) |
| Gradient Boosting | 0.662 ± 0.04 (0.625-0.698) | 0.022 ± 0.0031 (0.0051-0.0106) | 0.634 ± 0.04 (0.601-0.666) | 0.637 ± 0.05 (0.594-0.680) |
| AdaBoost | 0.657 ± 0.03 (0.630-0.684) | 0.007 ± 0 (0.0072-0.0072) | 0.625 ± 0.02 (0.609-0.641) | 0.635 ± 0.03 (0.605-0.665) |
| Random Forest | 0.545 ± 0.02 (0.525-0.565) | 0.008 ± 0.0002 (0.0073-0.0077) | 0.534 ± 0.03 (0.508-0.559) | 0.549 ± 0.05 (0.505-0.593) |
| AutoPrognosis | 0.679 ± 0.04 (0.642-0.716) | 0.007 ± 0.0010 (0.0058-0.0075) | 0.660 ± 0.02 (0.646-0.674) | 0.657 ± 0.04 (0.620-0.693) |
All values reported as mean ± standard deviation with (95% confidence interval).