Table 3. Testing model results from classification random forests for within- and between-participant analyses for varying levels of sleep deprivation and WR classifications.
Model accuracy (lower to upper 95% CI), area under the receiver operating curve (AUC, lower to upper 95% CI), negative and positive prediction accuracy (NPV and PPV%), specificity (SP%), and sensitivity (SN%) are displayed.
| X | Time grouping (hours) | Accuracy (%) (lower–upper) | AUC (%) (lower–upper) | SP (%) | SN (%) | NPV (%) | PPV (%) |
|---|---|---|---|---|---|---|---|
| Within participant | 0–16 vs. 18–38 | 92.4 (87.8–95.7) | 98 (96.6–99.4) | 91.7 | 93.0 | 90.6 | 93.8 |
| 0–16 vs. 20–38 | 93.6 (89.1–96.7) | 98.9 (98.1–99.8) | 92.9 | 94.2 | 92.9 | 94.2 | |
| 0–16 vs. 22–38 | 93.8 (89.2–96.9) | 98.8 (97.9–99.8) | 93.0 | 94.6 | 94.1 | 93.5 | |
| 0–18 vs. 24–38 | 96.6 (92.8–98.8) | 99.1 (98.1–100) | 98.9 | 94.3 | 94.7 | 98.8 | |
| 0–20 vs. 24–38 | 92.1 (87.2–95.5) | 97.9 (96.3–99.5) | 95.0 | 88.8 | 90.5 | 94.0 | |
| 0–23 vs. 24–38 | 90.4 (85.4–94.1) | 96.5 (94.2–98.7) | 90.6 | 90.1 | 93.0 | 86.9 | |
| Between participants | 0–16 vs. 18–38 | 70.2 (63.3–76.5) | 84.7 (79.4–90) | 64.8 | 74.5 | 67.1 | 72.6 |
| 0–16 vs. 20–38 | 72.3 (65.4–78.6) | 85.8 (80.6–91) | 67.7 | 76.8 | 74.1 | 70.9 | |
| 0–16 vs. 22–38 | 75.3 (68.3–81.4) | 87.6 (82.7–92.5) | 71.1 | 80.2 | 81.2 | 69.9 | |
| 0–18 vs. 24–38 | 78.2 (71.4–84) | 88.4 (83.7–93.1) | 76.9 | 80.0 | 84.2 | 71.4 | |
| 0–20 vs. 24–38 | 77.2 (70.6–83) | 88.1 (83.5–92.7) | 78.7 | 75.3 | 81.0 | 72.6 | |
| 0–23 vs. 24–38 | 76.8 (70.3–82.5) | 87.4 (82.8–92.1) | 79.3 | 73.2 | 80.7 | 71.4 |
Note: For training models tested above, tree depth ranged from 27 to 73 nodes