SleepExplain: Explainable Non-Rapid Eye Movement and Rapid Eye Movement Sleep Stage Classification from EEG Signal

Researchers have developed SleepExplain, a machine learning model that classifies sleep stages (NREM and REM) from EEG signals with 94.30% accuracy using XGBoost, while employing SHAP explainability techniques to make predictions interpretable. This advancement bridges clinical diagnostics and AI transparency, addressing a critical need in sleep disorder diagnosis where understanding model reasoning is as important as accuracy.

Sleep stage classification represents a foundational challenge in clinical neurology, where misdiagnosis can delay treatment for conditions ranging from sleep apnea to narcolepsy. The SleepExplain model achieves state-of-the-art performance by leveraging ensemble methods—Random Forest (92.54%), Gradient Boosting (94.25%), and XGBoost (94.30%)—demonstrating that combining multiple algorithms yields superior results compared to single-model approaches. The distinction lies not merely in accuracy but in the integration of explainability through SHAP values, a game-theoretic framework that deconstructs how each EEG feature contributes to classification decisions.

This development reflects a broader industry shift toward explainable AI (XAI) in healthcare, where regulatory bodies increasingly demand transparent decision-making pathways. Sleep medicine has traditionally relied on polysomnography technicians manually scoring EEG traces—a labor-intensive process prone to subjective variation. Automating this workflow while maintaining interpretability addresses clinician skepticism toward black-box neural networks in high-stakes diagnostic settings.

The market implications extend across digital health platforms, sleep clinic operators, and wearable manufacturers seeking FDA-compliant automated sleep staging. Hospitals can reduce diagnostic costs and turnaround times, while telemedicine providers gain tools for remote sleep monitoring. However, implementation requires clinical validation across diverse patient populations and integration with existing EEG infrastructure.

Looking forward, the challenge involves deploying such models in resource-constrained settings and ensuring equitable performance across demographic groups. Real-world adoption depends on regulatory approval, interoperability with legacy systems, and clinician acceptance—factors that transcend raw accuracy metrics.

• →XGBoost achieved 94.30% accuracy in NREM/REM classification, surpassing traditional ensemble methods by 1-2%.
• →SHAP explainability framework enables clinicians to understand which EEG features drive each sleep stage prediction.
• →Automated sleep staging could reduce manual scoring burden and diagnostic timelines in clinical settings.
• →Model performance depends on prospective validation across diverse patient populations and sleep disorders.
• →Integration with digital health platforms requires FDA compliance and compatibility with existing polysomnography infrastructure.