TABLE 3.
Performances of each training strategy and statistical differences. For each dataset, average Kappa agreement and the percentage of subjects above the threshold of substantial agreement (κ > .6) is given, followed by the group statistic and post hoc analysis between training strategies if significant. For the healthy datasets, no group statistic is shown since comparison is between two training strategies. Training-from-scratch is abbreviated to “scratch”.
| Dataset | N | Method | Kappa Mean ± SD |
Above substantial agreement (%) | Repeated measures ANOVA: p-value, effect size |
Post-hoc comparison between |
Paired samples t-test: p-value, effect size |
|---|---|---|---|---|---|---|---|
| Healthy F3-M2 |
60 | Pre-train | κ = .84 ± .06 | 98% | Pre-train & Scratch | p < .001, d = .27 | |
| Scratch | κ = .82 ± .08 | 97% | |||||
| Healthy F3-F4 |
60 | Pre-train | κ = .53 ± .16 | 35% | Pre-train & Scratch | p < .001, d = 1.87 | |
| Scratch | κ = .77 ± .09 | 93% | |||||
| OSA F3-M2 |
60 | Pre-train | κ = .74 ± .10 | 93% | p = .26, ηp 2 = .002 | ||
| Scratch | κ = .74 ± .09 | 95% | |||||
| Fine-tune | κ = .75 ± .09 | 95% | |||||
| OSA F3-F4 |
60 | Pre-train | κ = .53 ± .17 | 40% | p < .001, ηp 2 = .29 | Pre-train & Scratch | p < .001, d = 1.26 |
| Scratch | κ = .70 ± .09 | 87% | Pre-train & Fine-tune | p < .001, d = 1.21 | |||
| Fine-tune | κ = .70 ± .11 | 83% | Scratch & Fine-tune | p = .84, d = .02 | |||
| Insomnia F3-M2 |
60 | Pre-train | κ = .77 ± .09 | 98% | p = .27, ηp 2 = .003 | ||
| Scratch | κ = .77 ± .07 | 98% | |||||
| Fine-tune | κ = .78 ± .07 | 100% | |||||
| Insomnia F3-F4 |
60 | Pre-train | κ = .56 ± .13 | 43% | p < .001, ηp 2 = .23 | Pre-train & Scratch | p < .001, d = 1.39 |
| Scratch | κ = .73 ± .10 | 93% | Pre-train & Fine-tune | p < .001, d = 1.42 | |||
| Fine-tune | κ = .73 ± .11 | 92% | Scratch & Fine-tune | p = .53, d = .04 | |||
| RBD F3-M2 |
22 | Pre-train | κ = .60 ± .13 | 50% | p = .002, ηp 2 = .11 | Pre-train & Scratch | p = .43, d = .18 |
| Scratch | κ = .58 ± .12 | 45% | Pre-train & Fine-tune | p = .001, d = .65 | |||
| Fine-tune | κ = .67 ± .08 | 91% | Scratch & Fine-tune | p < .001, d = .89 | |||
| RBD F3-F4 |
22 | Pre-train | κ = .37 ± .18 | 14% | p < .001, ηp 2 = .31 | Pre-train & Scratch | p < .001, d = 1.06 |
| Scratch | κ = .54 ± .13 | 36% | Pre-train & Fine-tune | p < .001, d = 1.51 | |||
| Fine-tune | κ = .60 ± .11 | 55% | Scratch & Fine-tune | p = .01, d = .51 |