Enhancing Current Cardiorespiratory-based Approaches of Sleep Stage Classification by Temporal Feature Stacking

Area of Science:

• Biomedical Engineering
• Sleep Medicine
• Machine Learning

Background:

• Accurate sleep stage classification is crucial for diagnosing sleep disorders.
• Limited data and feature sets pose challenges in developing robust cardiorespiratory-based sleep analysis systems.
• Existing methods often struggle to incorporate temporal dynamics effectively.

Purpose of the Study:

• To present a generic method for enhancing sleep stage classification performance.
• To incorporate temporal information into cardiorespiratory-based sleep analysis.
• To improve classification accuracy using limited feature sets and data.

Main Methods:

• Utilized a random forest classification algorithm.
• Extracted features from long-term home monitoring for sleep analysis.
• Employed temporal feature stacking (pre- and post-epoch) to enhance classification.

Main Results:

• Significant improvements in Cohen's kappa (κ) and accuracy were observed.
• Performance gains were demonstrated across 3, 4, and 5 sleep stage classifications.
• Optimal stacking duration (30s-30min) and combination methods were analyzed.

Conclusions:

• The proposed temporal feature stacking method generically enhances classification performance for cardiorespiratory-based sleep analysis.
• This approach is effective for small-scale datasets and classification problems with temporal dependencies.
• The method offers a valuable tool for improving sleep analysis using random forest classifiers.