June 2, 2025Open Access

Interpretable feature-based machine learning for automatic sleep detection using photoplethysmography

Q: What is the clinical evidence from this study?

Study design: Other. Population: Sleep disorders (n=84). Intervention: Random Forest model with ADASYN balancing and 35 features (PPG-based sleep staging) vs. Unbalanced dataset model. Primary outcome: F1 score for sleep-wake classification (20-fold cross-validation).

Key Result

A Random Forest model using photoplethysmography with ADASYN balancing and 35 features achieved an F1 score of 81.40%, with 88.57% sensitivity for sleep and 71.31% specificity for wake detection.

Structured PICO

Does a feature-based machine learning model using photoplethysmography accurately classify sleep-wake states?

Population

84 participants from the Cyclic Alternating Pattern Sleep Database (85,542 epochs)

Intervention

Random Forest machine learning model using features extracted from photoplethysmography (PPG) data, with ADASYN balancing and 35 features

Comparator

Unbalanced dataset model (75 features)

Outcome

Sleep-wake classification performance (F1 score, sensitivity, specificity)surrogate

A feature-based Random Forest model using PPG data with ADASYN balancing achieved an F1 score of 81.40% for sleep-wake classification, offering a potential alternative to polysomnography for wearable devices.

Main Result

Absolute Event Rate: 81.4% vs 89.05%

Limitations

Initial model selection lacked prior feature selection and hyperparameter tuning.
Noise reduction methods could be more robust, as excessive noise filtering might limit applicability to real-world wearable data.
Balancing the training set while maintaining an untouched test set would further validate the model's effectiveness.
The study focused only on epoch-by-epoch binary sleep-wake classification rather than multi-class sleep staging.
Adapting the model for wearable use remains a challenge due to lower sampling rates, missing data, and computational constraints.

Abstract

Sleep disorders affect millions globally, leading to serious health issues. Accurate sleep-wake classification is essential for diagnosis and management. While polysomnography is the gold standard, it is costly and invasive; photoplethysmography (PPG) offers a viable alternative. Using the Cyclic Alternating Pattern Sleep Database (84 participants, 85,542 epochs), we extracted 330 features and reduced dimensionality via statistical tests and the SelectFromModel method. To address class imbalance, we applied Adaptive Synthetic (ADASYN) sampling. A Random Forest model, validated with 20-fold cross-validation on the unbalanced dataset (75 features), achieved an F1 score of 89.05% but struggled with wake detection. With ADASYN balancing and 35 features, it achieved 88.57% sensitivity (sleep) and 71.31% specificity (wake), with an F1 score of 81.40%. This feature-based approach improves PPG-based sleep classification, supporting clinical adoption and integration into wearable devices for remote sleep monitoring.

Bookmark

View Full Paper

Cite This Study

Markov et al. (Mon,) conducted a other in Sleep disorders (n=84). Random Forest model with ADASYN balancing and 35 features (PPG-based sleep staging) vs. Unbalanced dataset model was evaluated on F1 score for sleep-wake classification (20-fold cross-validation). A Random Forest model using photoplethysmography with ADASYN balancing and 35 features achieved an F1 score of 81.40%, with 88.57% sensitivity for sleep and 71.31% specificity for wake detection.

synapsesocial.com/papers/6a10df4339dd87f6d0ee7a7f https://doi.org/https://doi.org/10.1038/s44328-025-00041-2

Also Consider

Synapse has enriched 5 closely related papers on similar clinical questions. Consider them for comparative context:

Bookmark

View Full Paper