Random Forest Model for Sleep Stage Classification Using Approximate Entropy from RR Intervals

Santiago Ismael Flores-Chavez1, Antonio Gabriel Ravelo-Garcı́a2, Miguel Vizcardo1
1Universidad Nacional de San Agustin de Arequipa, 2Institute for Technological Development and Innovation in Communications, Universidad de Las Palmas de Gran Canaria, 35017 Las Palmas de Gran Canaria, Spain Interactive Technologies Institute (ITI/LARSyS and ARDITI), 9020-105 Funchal, Portugal


Abstract

Abstract: Sleep stage classification plays an important role in the diagnosis of sleep disorders. This study proposes a classification model based on a Random Forest algorithm trained with features derived from Approximate Entropy (ApEn) of RR intervals to classify different sleep stages. Various embedding dimensions and time windows were evaluated to extract meaningful features. The model demonstrated promising results, suggesting that entropy-based analysis of heart rate variability contains valuable information for sleep stage discrimination. These findings indicate potential applications for simplified sleep monitoring using cardiac signals. Method: Data from subjects with annotated sleep stages were analyzed using RR interval time series. ApEn was computed across multiple embedding dimensions and time windows. Feature extraction involved calculating statistical measures from these entropy values. A Random Forest classifier was trained on a portion of the data and evaluated on a held-out test set. Results: The Random Forest model achieved satisfactory accuracy in distinguishing between sleep stages. Certain embedding dimensions and time window selections appeared more effective for specific classification tasks. Entropy patterns showed different characteristics across various sleep states, aligning with existing literature on physiological dynamics during sleep. Conclusions: This study supports the use of Random Forest classification with ApEn features for sleep stage identification. The approach demonstrates that entropy-based analysis of cardiac intervals can capture meaningful aspects of sleep architecture. Clinical applications could include less invasive sleep monitoring alternatives. Further research with larger populations would be beneficial to validate these findings.