Sleep classification from wrist-worn accelerometer data using Random Forests

Accurate and low cost sleep measurement tools are needed in both clinical and epidemiological research. To this end, wearable accelerometers are widely used as they are both low in price and provide reasonably accurate estimates of movement. Techniques to classify sleep from the high resolution accelerometer data primarily rely on heuristic algorithms. In this paper, we explore the potential of detecting sleep using Random forests. Models were trained using data from three different studies where 158 participants (78 with sleep disorder and 80 good healthy sleepers) wore an  accelerometer on their wrist during a one night Polysomnography recording in the clinic. The Random forests were able to distinguish  sleep-wake states at a 30-second resolution with a F1 score of 73.93\% on a previously unseen test set of 24 participants.

These Random forest models have been made open-source to aid further research in sleep analysis. The models can be used with GGIR as described in https://github.com/wadpac/SleepStageClassification/tree/master/ggir_ext . These models were trained with scikit-learn v0.22.1 and we recommend using the same version for using these models. If used with other versions, there might be a warning indicating incompatibility.