Combining deep belief networks and bidirectional long short-term memory: Case study: Sleep stage classification

This paper proposes a new combination of Deep Belief Networks (DBN) and Bidirectional Long Short-Term Memory (Bi-LSTM) for Sleep Stage Classification. Tests were performed using sleep stages of 25 patients with sleep disorders. The recording comes from electroencephalography (EEG), electromyography (EMG), and electrooculography (EOG) represented in signal form. All three of these signals processed and extracted to produce 28 features. The next stage, DBN Bi-LSTM is applied. The analysis of this combination compared with the DBN, DBN HMM (Hidden Markov Models), and Bi-LSTM. The results obtained that DBN Bi-LSTM is the best based on precision, recall, and F1 score.