On average, sleep/wake classification accuracies were reported to be 87.2% based on 53 assessed devices. There was no significant difference in accuracies between devices using only accelerometer data (86.7%, d = 28) and devices using both PPG and accelerometer data (87.8%, d = 22), as determined by a t-test (significance threshold p < 0.05). All reported accuracies ranged from 79% to 96%, except for Kanady et al.’s study28, which reported lower values of 54% and 64%. This difference can be attributed to their 24-hour measurement, which had a higher wake-to-sleep ratio compared to overnight measurements in other studies. Therefore, these accuracies reflect the generally poor performance of sleep classifiers in detecting wake. The average accuracy for 3-stage classification (wake vs. NREM vs. REM) was 69.7% (d = 3), and for 4-stage classification (wake vs. light vs. deep vs. REM), it was 65.2% (d = 9).
Evaluating reliability in wearable devices for sleep staging · HackerTrans
On average, sleep/wake classification accuracies were reported to be 87.2% based on 53 assessed devices. There was no significant difference in accuracies between devices using only accelerometer data (86.7%, d = 28) and devices using both PPG and accelerometer data (87.8%, d = 22), as determined by a t-test (significance threshold p < 0.05). All reported accuracies ranged from 79% to 96%, except for Kanady et al.’s study28, which reported lower values of 54% and 64%. This difference can be attributed to their 24-hour measurement, which had a higher wake-to-sleep ratio compared to overnight measurements in other studies. Therefore, these accuracies reflect the generally poor performance of sleep classifiers in detecting wake. The average accuracy for 3-stage classification (wake vs. NREM vs. REM) was 69.7% (d = 3), and for 4-stage classification (wake vs. light vs. deep vs. REM), it was 65.2% (d = 9).