Table 2.
End-to-end performance of the best-performing model predicting the SARA gait score on the ordinal scale [0, 1, 2, 3, 4] in comparison with the human performance
| Model | Prec. ↑ | Rec. ↑ | F1 ↑ |
|---|---|---|---|
| tsfresh+XGBoostH41 | 74.87 | 67.17 | 63.99 |
| Human Baseline | 73.03 | 66.68 | 60.57 |
Results of the classification experiment across all patient SARA gait score classes [0,1,2,3,4] evaluated on the subset of the 41 cases, for which the human baseline of consensus video ratings of 3 experts were available. Only the tsfresh+XGBoost model surpassed the human baseline. All scores are presented in %. Arrows indicate the favorable outcome; for all considered metrics, higher values are favorable.
Prec. precision, Rec. recall, F1 F1-score.