| Â | RF | LR |
|---|
% train | Lit | TS | Sign | Lit | TS | Sign |
|---|
80 | 0.73±0.08 | 0.75±0.07 | 12.1% | 0.67±0.10 | 0.72±0.08 | 21.8% |
50 | 0.71±0.05 | 0.73±0.04 | 18.7% | 0.67±0.06 | 0.71±0.05 | 35.8% |
30 | 0.70±0.04 | 0.72±0.04 | 24.7% | 0.65±0.05 | 0.69±0.04 | 37.9% |
20 | 0.68±0.04 | 0.71±0.04 | 30.1% | 0.63±0.06 | 0.67±0.05 | 42.2% |
- The leftmost column indicates what percentage of the dataset was used for training. Results are shown for the classifier using just latencies, EDSS at T0 and age (Lit), and for the classifier trained on these features + additional TS features (TS). RF (Random Forest) and LR (Logistic Regression) indicate the classifier that was used. The sign column indicates the percentage of splits with a significant improvement, according to the DeLong test. These results are shown graphically in Fig. 4. The values after ± indicate the standard deviations
