Sleeping Stages Scoring Automation by Utilising Time-Frequency Spectra and Convolutional Neural Networks

Automated sleep stage rating using EEG data and convolutional neural networks (CNNs). Through time-frequency approaches, we examine these signals that exhibit frequency as well as amplitude variations. Using raw EEG data and associated time-frequency modifications, the research investigates several CNN designs, including 1D-CNN, SWT-CNN, and STFT-CNN. These types of systems excel at integrating pre-treatment procedures alongside the classification stage, which makes them ideal for power- and storage-constrained handheld devices. The suggested models show great accuracy in identifying the different phases of sleep when evaluated on EEG samples. To enhance extracting features, each CNN model makes use of its own unique set of preliminary processing methods, including sustained wavelet transform (SWT) and short-time Fourier transform (STFT). The models generated by CNN outperformed human scorers, proving their usefulness for automated, real-time tracking of sleep in a variety of contexts. The research highlights the effectiveness of less complicated approaches for biosignal analysing, especially in healthy persons. It suggests that subsequent studies may focus on fine-tuning to account for atypical routines of sleep. Compared to existing models, the suggested technique yields a remarkable $98 \%$ accuracy. The accuracy of Deep-CNN is the lowest, whereas CNN-LSTM (Long Short-Term Memory) is somewhat better. When compared to the suggested approach, the Ensemble SVM performs better, while the Multitask CNN is the most comparable.