Clustering-Based Detection of Driver Drowsiness with Ear-EEG

Eray Mutlu; Matteo Saibene; Jacopo de Araujo; Esben Ahrens; Sadasivan Puthusserypady

doi:10.1109/BCI69045.2026.11435106

Clustering-Based Detection of Driver Drowsiness with Ear-EEG

T&W Engineering A/S

Research output: Chapter in Book/Report/Conference proceeding › Article in proceedings › Research › peer-review

Abstract

Fatigue remains one of the leading factors in road accidents, necessitating effective solutions for driver drowsiness detection. This study explores the potential of in-ear electroencephalogram (EEG) combined with unsupervised learning to detect drowsiness during simulated driving. Data from 17 sessions involving 9 participants were analyzed using 8 spectral features, including novel composite metrics r1 and r2. Three clustering algorithms: K-means, Gaussian Mixture Models (GMM), and DBSCAN—were employed to classify cognitive states without labeled data. The K-means yielded the highest performance (Silhouette Score: 0.93), with r1 and r2 outperforming traditional band ratios in distinguishing alert and drowsy states. These results demonstrate that ear-EEG, combined with unsupervised clustering, offers a viable pathway toward real-time, label-free fatigue detection systems with potential for practical deployment.

Original language	English
Title of host publication	Proceedings of the 14^th International Conference on Brain-Computer Interface
Number of pages	6
Publisher	IEEE
Publication date	2026
Article number	11435106
ISBN (Print)	979-8-3315-7928-9
DOIs	https://doi.org/10.1109/BCI69045.2026.11435106
Publication status	Published - 2026
Event	14^thIEEE International Winter Conference on Brain-Computer Interface Conference 2026 - High1 Resort, Gangwon, Korea, Republic of Duration: 23 Feb 2026 → 25 Feb 2026

Conference

Conference	14^thIEEE International Winter Conference on Brain-Computer Interface Conference 2026
Location	High1 Resort
Country/Territory	Korea, Republic of
City	Gangwon
Period	23/02/2026 → 25/02/2026

Keywords

Measurement
Clustering algorithms
Feature extraction
Fatigue
Real-time systems
Brain-computer interfaces
Safety
Unsupervised learning
Spectrogram
Vehicles

Access to Document

10.1109/BCI69045.2026.11435106

OpenUrl availability

Check availability in DTU Findit

Cite this

@inproceedings{c6cab7b2cfe146aa841e68984bd2c7ca,

title = "Clustering-Based Detection of Driver Drowsiness with Ear-EEG",

abstract = "Fatigue remains one of the leading factors in road accidents, necessitating effective solutions for driver drowsiness detection. This study explores the potential of in-ear electroencephalogram (EEG) combined with unsupervised learning to detect drowsiness during simulated driving. Data from 17 sessions involving 9 participants were analyzed using 8 spectral features, including novel composite metrics r1 and r2. Three clustering algorithms: K-means, Gaussian Mixture Models (GMM), and DBSCAN—were employed to classify cognitive states without labeled data. The K-means yielded the highest performance (Silhouette Score: 0.93), with r1 and r2 outperforming traditional band ratios in distinguishing alert and drowsy states. These results demonstrate that ear-EEG, combined with unsupervised clustering, offers a viable pathway toward real-time, label-free fatigue detection systems with potential for practical deployment.",

keywords = "Measurement, Clustering algorithms, Feature extraction, Fatigue, Real-time systems, Brain-computer interfaces, Safety, Unsupervised learning, Spectrogram, Vehicles",

author = "Eray Mutlu and Matteo Saibene and Araujo, \{Jacopo de\} and Esben Ahrens and Sadasivan Puthusserypady",

year = "2026",

doi = "10.1109/BCI69045.2026.11435106",

language = "English",

isbn = "979-8-3315-7928-9",

booktitle = "Proceedings of the 14th International Conference on Brain-Computer Interface",

publisher = "IEEE",

address = "United States",

note = "14<sup>th </sup>IEEE International Winter Conference on Brain-Computer Interface Conference 2026, bci2026 ; Conference date: 23-02-2026 Through 25-02-2026",

}

Mutlu, E, Saibene, M , Araujo, JD, Ahrens, E & Puthusserypady, S 2026, Clustering-Based Detection of Driver Drowsiness with Ear-EEG. in Proceedings of the 14^th International Conference on Brain-Computer Interface., 11435106, IEEE, 14^thIEEE International Winter Conference on Brain-Computer Interface Conference 2026, Gangwon, Korea, Republic of, 23/02/2026. https://doi.org/10.1109/BCI69045.2026.11435106

TY - GEN

T1 - Clustering-Based Detection of Driver Drowsiness with Ear-EEG

AU - Mutlu, Eray

AU - Saibene, Matteo

AU - Araujo, Jacopo de

AU - Ahrens, Esben

AU - Puthusserypady, Sadasivan

PY - 2026

Y1 - 2026

N2 - Fatigue remains one of the leading factors in road accidents, necessitating effective solutions for driver drowsiness detection. This study explores the potential of in-ear electroencephalogram (EEG) combined with unsupervised learning to detect drowsiness during simulated driving. Data from 17 sessions involving 9 participants were analyzed using 8 spectral features, including novel composite metrics r1 and r2. Three clustering algorithms: K-means, Gaussian Mixture Models (GMM), and DBSCAN—were employed to classify cognitive states without labeled data. The K-means yielded the highest performance (Silhouette Score: 0.93), with r1 and r2 outperforming traditional band ratios in distinguishing alert and drowsy states. These results demonstrate that ear-EEG, combined with unsupervised clustering, offers a viable pathway toward real-time, label-free fatigue detection systems with potential for practical deployment.

AB - Fatigue remains one of the leading factors in road accidents, necessitating effective solutions for driver drowsiness detection. This study explores the potential of in-ear electroencephalogram (EEG) combined with unsupervised learning to detect drowsiness during simulated driving. Data from 17 sessions involving 9 participants were analyzed using 8 spectral features, including novel composite metrics r1 and r2. Three clustering algorithms: K-means, Gaussian Mixture Models (GMM), and DBSCAN—were employed to classify cognitive states without labeled data. The K-means yielded the highest performance (Silhouette Score: 0.93), with r1 and r2 outperforming traditional band ratios in distinguishing alert and drowsy states. These results demonstrate that ear-EEG, combined with unsupervised clustering, offers a viable pathway toward real-time, label-free fatigue detection systems with potential for practical deployment.

KW - Measurement

KW - Clustering algorithms

KW - Feature extraction

KW - Fatigue

KW - Real-time systems

KW - Brain-computer interfaces

KW - Safety

KW - Unsupervised learning

KW - Spectrogram

KW - Vehicles

U2 - 10.1109/BCI69045.2026.11435106

DO - 10.1109/BCI69045.2026.11435106

M3 - Article in proceedings

SN - 979-8-3315-7928-9

BT - Proceedings of the 14th International Conference on Brain-Computer Interface

PB - IEEE

T2 - 14<sup>th </sup>IEEE International Winter Conference on Brain-Computer Interface Conference 2026

Y2 - 23 February 2026 through 25 February 2026

ER -

Clustering-Based Detection of Driver Drowsiness with Ear-EEG

Abstract

Conference

Keywords

Access to Document

OpenUrl availability

Fingerprint

Cite this