Hybrid Deep Learning Model Based on Transformer Encoder for Sleep Stages Classification

Abstract

Sleep is the cornerstone of overall health, and the process of sleep staging involves classifying sleep data into specific stages. Key signals such as EEG, EOG, and EMG are useful in analysing and categorizing sleep data, but it is a complex and time-consuming task. This paper focuses on designing a hybrid deep learning model to accurately classify sleep data using the Sleep Heart Study (SHHS) dataset. Considering that sleep signals show similar temporal patterns to the time series data, we also use transform encoders to extract essential features and facilitate the discrimination of sleep stages. By leveraging the power of transform encoders to capture crucial temporal patterns, we have successfully enhanced the classification of sleep data into five stages. Through comprehensive evaluation using various criteria, we measure the performance of our model and compare it with cutting-edge methods. The results show a significant accuracy of 0.883 and 0.836 for accuracy and Cohen's kappa, respectively, confirming the effectiveness of our approach. The results also highlight the robustness and efficiency of our approach in accurately diagnosing sleep states, ultimately contributing to the advancement of sleep analysis and overall health monitoring.