A probability model using EEG relative power features successfully separated three distinct clusters of alertness (wakefulness, drowsiness, and sleep) at high resolution with a detection accuracy of 93.21% and mean silhouette value of 0.74, suggesting potential future applications in drowsiness detection.
Key Findings
Methods
A high-resolution EEG-based probability model successfully identified three distinct clusters of alertness: wakefulness, drowsiness, and sleep.
EEG signals were collected from 53 subjects during overnight sleep studies
Relative power features were extracted from EEG to develop a model yielding likelihood of wakefulness for each 3-second EEG segment
The model yielded the likelihood of wakefulness as a continuous probability rather than a binary classification
Three clusters were identified and validated using statistical analyses, cluster quality evaluation, and graphical analysis
Results
The three discovered clusters were compact and well-separated, as indicated by strong cluster quality metrics.
Mean silhouette value on test data was 0.74, indicating good cluster separation
Mean Davies-Bouldin index value was 0.43, indicating compact clusters (lower values indicate better separation)
Detection accuracy based on silhouette values was 93.21%
One-way repeated measures ANOVA confirmed feature values were significantly different among the three clusters (p < .0001)
Results
The proposed method was able to detect short episodes of wakefulness, drowsiness, and sleep with high accuracy in overnight polysomnography data.
Analysis was performed at 3-second EEG segment resolution, which is higher resolution than conventional 30-second epoch scoring
The method captured dynamics of wakefulness/sleep transition as a gradual and continuous process rather than instantaneous
The model successfully separated three distinct cases of alertness in overnight recordings
Conventional sleep scoring is described as instantaneous and scored subjectively at low resolution, which this method improves upon
Background
Wakefulness/sleep transition has conventionally been considered instantaneous and scored subjectively at low resolution, despite being a gradual and continuous process.
Standard polysomnography uses 30-second epochs for sleep staging, which the authors characterize as low resolution
The authors identify a gap between the biological reality of gradual sleep onset and the instantaneous binary scoring used in practice
Capturing the dynamics of sleep onset process is described as fundamental to sleep medicine and circadian neurobiology
The study presents a proof-of-concept for quantitatively capturing wakefulness to sleep transition dynamics
Conclusions
This proof-of-concept study suggests potential future applications in drowsiness detection pending validation in relevant contexts such as driving simulators and workplace environments.
The study was conducted using overnight polysomnography data, not real-world drowsiness contexts
Authors explicitly note the need for validation in driving simulators and workplace environments before practical application
The model is described as efficient, high-resolution, and reliable for quantitatively capturing wakefulness/sleep transition dynamics
The study used data from 53 subjects in a sleep study setting
What This Means
This research suggests that a new computer model can detect drowsiness more precisely than current methods by analyzing brain wave patterns (EEG) at very fine time intervals of just 3 seconds. Rather than simply labeling someone as 'awake' or 'asleep' — as standard sleep testing does in 30-second blocks — this model assigns a continuous probability score reflecting how alert a person is, and groups EEG segments into three states: wakefulness, drowsiness, and sleep. The model was tested on overnight sleep recordings from 53 people and achieved an accuracy of over 93%, with strong statistical evidence that the three states were genuinely distinct from one another.
The key innovation is capturing the gradual transition between wakefulness and sleep, which is biologically a continuous process but has traditionally been treated as an instantaneous switch in clinical practice. By identifying a 'drowsiness' state that sits between full wakefulness and sleep, the model can potentially detect the early stages of falling asleep with greater precision and in shorter time windows than existing methods.
This research suggests the technology could eventually be applied in real-world safety settings, such as monitoring drivers for drowsiness or detecting inattention in workplace environments. However, the authors caution that this is a proof-of-concept study conducted in a controlled sleep lab, and further validation in those real-world contexts would be needed before practical use.
Hassan A, Kabir M, Saha S, Keshavarz B, Yadollahi A. (2026). Development of a probability model for high-resolution drowsiness detection using electroencephalogram.. Sleep medicine. https://doi.org/10.1016/j.sleep.2025.108733