Table 1.
Scoring performance on large-scale and small-scale datasets.
| Scoring Methods | WAKE | Non-REM | REM | Acc. | κ stat. | |||
|---|---|---|---|---|---|---|---|---|
| Rec. | Prec. | Rec. | Prec. | Rec. | Prec. | |||
| on large-scale dataset of 4,200 mice | ||||||||
| MC-SleepNet | 98.1% | 97.8% | 95.8% | 97.3% | 80.1% | 90.1% | 96.7% | 0.94 |
| MC-SleepNet + rescoring | 98.6% | 97.0% | 94.4% | 97.8% | 89.5% | 77.9% | 96.3% | 0.94 |
| Random forest | 95.9% | 95.0% | 94.0% | 94.1% | 76.9% | 84.0% | 94.1% | 0.89 |
| MC-SleepNet (noisy records) | 94.8% | 98.5% | 97.5% | 92.5% | 75.7% | 86.8% | 95.0% | 0.91 |
| on small-scale dataset of 14 mice | ||||||||
| MC-SleepNet | 99.3% | 98.6% | 93.9% | 99.6% | 99.5% | 62.6% | 96.4% | 0.94 |
| Random forest | 95.9% | 92.6% | 94.3% | 95.6% | 73.9% | 88.9% | 94.0% | 0.89 |
| FASTER2 | 89.5% | — | 94.2% | — | 78.4% | — | 91.1% | — |
| MASC4 | 95.5% | 97.3% | 94.7% | 96.9% | 94.0% | 65.8% | 95.0% | 0.91 |
| LSTM model5 | 96.2% | 94.8% | 95.1% | 95.8% | 82.1% | 85.2% | 94.9% | 0.91 |