Fig 1.
Overview over the different datasets and how they were used in this study.
For each dataset, the key demographic features, which sensors were used and the target variable on which the model was evaluated are shown.
Fig 2.
Validation results on the Tromsø Study accelerometer data subsample.
(A) Accuracies for the same criteria as for the ActiGraph/Oura/Polar dataset. (B) Predicted times in bed plotted against the annotated TiB. The selected model reached an accuracy of over 90% on the additional, unseen data from the Tromsø Study and up to 95.6% when removing unreasonably long predictions (outliers of less than one or more than 23 h of predicted time in bed).