Table 2.
Classification matrix and subset accuracy of machine learning models to predict one-year chiropractic service utilization, based on parameters from initial development phase
| Model/Class | Precision (%) | Recall (%) | F-measure (%) | Accuracy (%) |
|---|---|---|---|---|
| Gradient Boosted Classifier | ||||
| 1 Visit | 43.5 | 61.7 | 51.0 | |
| 2–3 Visits | 36.3 | 32.5 | 34.3 | |
| 4–6 Visits | 34.0 | 9.2 | 14.5 | |
| 7+ Visits | 46.2 | 57.8 | 51.0 | |
| Averagea | 40.3 | 42.1 | 39.1 | 42.1 |
| Stochastic Gradient Descent Classifier | ||||
| 1 Visit | 43.9 | 53.1 | 48.1 | |
| 2–3 Visits | 32.3 | 29.9 | 31.0 | |
| 4–6 Visits | 24.9 | 14.8 | 18.6 | |
| 7+ Visits | 43.4 | 51.2 | 47.0 | |
| Averagea | 36.8 | 38.6 | 37.2 | 38.6 |
| Support Vector Classifier | ||||
| 1 Visit | 42.6 | 60.8 | 50.1 | |
| 2–3 Visits | 35.3 | 26.1 | 30.0 | |
| 4–6 Visits | 32.0 | 11.3 | 16.7 | |
| 7+ Visits | 45.3 | 60.3 | 51.7 | |
| Averagea | 39.2 | 41.4 | 38.4 | 41.4 |
| Artificial Neural Network | ||||
| 1 Visit | 42.9 | 56.2 | 48.7 | |
| 2–3 Visits | 35.9 | 30.6 | 33.1 | |
| 4–6 Visits | 25.3 | 12.1 | 16.4 | |
| 7+ Visits | 45.1 | 55.9 | 49.9 | |
| Averagea | 38.0 | 40.3 | 38.2 | 40.3 |
aSupport-weighted average