Table 2.
Results of optimal training model performance and test prediction of each machine learning algorithm for multi-class classification.
| Classifier | Training model | Test prediction | Balanced accuracy by class | ||
|---|---|---|---|---|---|
| Accuracy, % | Overall accuracy, % (95% CI) |
Mild, % | Moderate, % | Severe, % | |
| Neural Network | 72.7 | 73.7 (68.1–78.8) | 83.6 | 66.3 | 84.3 |
| Support Vector Machines | 73.0 | 74.5 (68.9–79.5) | 82.9 | 68.5 | 81.4 |
| k-Nearest Neighbors | 71.6 | 75.6 (70.0–80.5) | 82.5 | 64.2 | 84.4 |
| Classification And Regression Tree | 73.6 | 70.8 (65.0–76.1) | 78.1 | 61.8 | 81.7 |
| Random Forest | 76.0 | 75.6 (70.0–80.5) | 83.6 | 70.5 | 81.9 |
| Stochastic Gradient Boosting | 74.7 | 73.4 (67.7–78.5) | 80.3 | 71.6 | 81.6 |
| eXtreme Gradient Boosting | 76.1 | 76.6 (71.2–81.5) | 83.6 | 71.8 | 83.5 |
CI, confidence interval.