Table 1.
Comparison of wake and sleep classification results (leave-one-subject-out cross validation—LOOCV) with video-based actigraphy.
| Metrics 1 | Feature Set | ||
|---|---|---|---|
| mACT | pSLP | mACT + pSLP | |
| True positive (TP) | 304 | 307 | 318 |
| False positive (FP) | 36 | 143 | 35 |
| False negative (FN) | 85 | 82 | 71 |
| True negative (TN) | 954 | 847 | 955 |
| Precision | 82.4% ± 18.3% | 61.0% ± 22.7% | 82.3% ± 16.6% |
| Sensitivity | 71.9% ± 24.4% | 82.8% ± 18.9% | 77.4% ± 22.7% |
| Specificity | 95.9% ± 2.6% | 84.8% ± 6.5% | 95.8% ± 2.6% |
| Accuracy | 91.3% ± 4.7% | 82.0% ± 7.9% | 92.0% ± 4.6% |
| Cohen’s kappa | 0.701 ± 0.227 | 0.544 ± 0.187 | 0.733 ± 0.204 |
1 For confusion matrix elements, results pooled over all epochs from all infants are presented; for the other metrics, mean ± standard deviation results across infants are presented. The highest accuracy and Cohen’s kappa are in bold.