Evaluation of the Effect of Signal Quality on Sleep Stage Classification Using Random Forest Algorithm

Manual scoring is commonly employed in polysomnography (PSG) to record, score, and interpret physiological sleep data. This method analyzes multiple signals to detect disorders but is hindered by its time-consuming, particularly given the typical 8-hour sleep period, and is susceptible to intra-scorer variability, resulting in inconsistent outcomes. Automated methods are essential to enhance efficiency, consistency, and accuracy in PSG scoring. This study evaluates the impact of data quality on sleep stage classification using the Random Forest method. PSG signals were preprocessed and analyzed in Matlab, with feature extraction in the time and frequency domains, including signal amplitude and power spectral density in delta and beta bands. Features were selected based on criteria for classifying Rapid Eye Movement (REM) and Non- Rapid Eye Movement (NREM) stages. Results showed a 2.07% improvement in REM sensitivity a 0.22% increase in NREM sensitivity for high-quality data. This underscores the importance of measuring data quality in sleep stage classification, especially in less controlled environments like hospitals or homes.