Deep Transfer Learning for Sleep Stages Classification by EEG Data

Sleep is one of the key factors of well-being affecting life quality and longevity, which makes sleep monitoring an important task. In this paper, we propose the transfer learning approach for sleep stages classification based on EEG recordings, aiming the analysis of the patient's sleep quality. Transfer learning makes training possible with a small amount of labeled data. Also, usually it takes less time to fine-tune a pretrained model. For the numerical experiments, we use an open PhysioNet Sleep-EDF dataset containing more than 150 EEG files with annotations, and recurrent autoencoder network TimeNet. The proposed approach provides the highest achievable binary classification accuracy when only 10% of labeled data used for training.