Table 3 Five-fold cross-validated performance for all models using dynamic accelerometric features. Values are expressed as mean (min–max) across folds.
Outcome (Dynamic) | Model | MSE (mean [min–max]) | \(R^2\) (mean [min–max]) |
|---|---|---|---|
Flamingo | Linear regression | 18.25 (16.70–19.90) | − 0.03 (− 0.20–0.11) |
Decision tree | 43.35 (39.00–48.20) | − 1.45 (− 1.70– − 1.20) | |
Random forest | 24.13 (21.50–27.80) | − 0.37 (− 0.55– − 0.18) | |
k-nearest neighbors | 23.90 (21.80–27.00) | − 0.36 (− 0.52– − 0.18) | |
Support vector regression | 25.40 (23.10–28.70) | − 0.42 (− 0.61– − 0.25) | |
Gradient boosting | 26.70 (24.00–30.10) | − 0.49 (− 0.68– − 0.31) | |
Balance beam | Linear regression | 12.50 (11.30–13.80) | 0.13 (0.01–0.25) |
Decision tree | 32.20 (29.00–35.40) | − 1.24 (− 1.50– − 0.95) | |
Random forest | 16.83 (15.00–18.90) | − 0.17 (− 0.35–0.02) | |
k-Nearest Neighbors | 16.40 (15.20–18.10) | − 0.15 (− 0.28–0.01) | |
Support vector regression | 17.50 (15.70–19.60) | − 0.19 (− 0.32–0.03) | |
Gradient boosting | 18.90 (17.00–21.00) | − 0.23 (− 0.38– − 0.06) |
