Table 2.
The model performance metrics of six different algorithms.
| Model | R2 | MAE | MSE | Accuracy-1 a | Accuracy-2 b |
|---|---|---|---|---|---|
| SVR | 0.676 | 3.868 | 26.071 | 76.79% | 62.50% |
| GBRT | 0.670 | 4.054 | 26.568 | 69.64% | 62.50% |
| RF | 0.656 | 4.410 | 27.683 | 62.50% | 48.21% |
| Bagging | 0.652 | 4.440 | 28.059 | 64.29% | 44.64% |
| Adaboost | 0.610 | 4.743 | 31.386 | 55.36% | 48.21% |
| XGBoost | 0.551 | 4.630 | 36.186 | 60.71% | 55.36% |
SVR, Support Vector Regression; GBRT, Gradient Boosted Regression Trees; RF, Random Forest; Bagging, Boostrap aggregating; Adaboost, Adaptive Boosting; XGBoost, eXtreme Gradient Boosting.
a
Absolute accuracy, the predict trough concentration was within ± 5 mg/l of the observed trough concentration.
b
Relative accuracy, the predict trough concentration was within ± 30% of the observed trough concentration.