Classification of Sleep Stages Using Single-Channel ECG Signals: A Comparative Analysis of Machine Learning and Deep Learning Methods

In this paper, we provide an extensive evaluation of machine learning (ML) and deep learning (DL) methods for automatic sleep stage classification using a single-channel electrocardiogram (ECG) signal. To explore ML methods, we extracted 21 heart rate variability (HRV) features using the R-peak detection. These features were used to train and compare the performance of various models such as support vector machine (SVM), decision tree (DT), random forest (RF), k-nearest neighbor (K-NN), and Naïve Bayes (NB). Furthermore, we analyzed the DL methods to overcome the manual feature extraction. This method involved feeding raw time-series ECG signal to a 1-dimensional convolutional neural network (CNN) and short-time Fourier transform (STFT) based spectrogram to a 2D CNN. The overall evaluation was performed using 15-subject data each from two publicly available datasets. To process the data for analysis, resampling, bandpass filtering, baseline correction, 30-second interval extraction from whole-night ECG recordings, and z-score normalization techniques were applied. We conducted experiments on one dataset and then replicated the experiments on the other dataset to assess the robustness and generalizability of the methods. Through a comparative analysis, the spectrogram-based 2D CNN emerged as the effective method with the highest test accuracy and k-score of 87.65% and 0.82 respectively. This study contributes to ECG-based sleep classification research, highlighting the effectiveness of deep learning for effective classification.