EEG signal classification for drowsiness detection using wavelet transform and support vector machine

Br. Pasaribu, Novie Theresia; Halim, Timotius; Ratnadewi, Ratnadewi; Prijono, Agus · 2021 · Crossref

DOI: 10.11591/ijai.v10.i2.pp501-509

archive: archived pipeline: cataloged verified

Get this paper ↗ (DOI — opens at the source; we link to it, we don't host it)

Summary

This study addresses the critical safety issue of driver drowsiness, which significantly increases the risk of traffic accidents due to reduced concentration. While various detection methods exist, including subjective scales and behavioral observations like eye aspect ratio (EAR), physiological methods using electroencephalogram (EEG) signals are considered the most objective. The research aims to develop an early drowsiness detection system by classifying EEG signals using wavelet transform for feature extraction and support vector machine (SVM) for classification. The system specifically analyzes Alpha, Beta, and Theta waves to distinguish between alert, relaxed, and light sleep states. The experimental design involved ten healthy respondents aged 19–26. Data was collected using a driving simulator, a camera for EAR calculation, and an Emotiv EPOC neuroheadset with 14 channels to record EEG signals. The protocol consisted of a baseline driving phase (Driving-1), a second driving phase (Driving-2), a 30-minute arithmetic stress task to induce fatigue, and a final driving phase (Driving-3). EEG signals from frontal (F3, F4, F7, F8) and occipital (O1, O2) channels were preprocessed and segmented into two-minute intervals. Drowsiness labels were assigned based on a modified EAR threshold, where an EAR below the threshold for 1.5 seconds indicated drowsiness. Feature extraction utilized Daubechies (Db4) wavelet transform at level 4, generating 66 coefficients per segment. These features were input into an SVM classifier, trained using 5-fold cross-validation. The results demonstrated that the quadratic kernel SVM achieved the highest training accuracy of 84.5%, outperforming linear (71.8%), cubic (52.3%), and various Gaussian kernels. In the testing phase, the system classified respondents into drowsy or awake categories based on whether at least two-thirds of the SVM outputs indicated drowsiness. During the Driving-2 process, seven respondents were classified as drowsy, while three remained awake. In the Driving-3 process, following the stress induction, six respondents were detected as drowsy and four as awake. The study confirms that the combination of wavelet transform and SVM effectively distinguishes between drowsy and alert states using EEG data correlated with behavioral EAR metrics. The significance of this work lies in its validation of a non-invasive, physiological approach to real-time drowsiness detection. By achieving high accuracy with a specific kernel configuration, the study provides a robust framework for integrating EEG-based monitoring into driver assistance systems. The findings suggest that combining physiological signals with behavioral cues enhances detection reliability. The authors conclude that while the current system operates on segmented time periods, future research should focus on optimizing the method for shorter, real-time detection windows to improve immediate driver safety interventions.

Provenance

The full processing record for this entry. Every stage of this paper's journey through the pipeline is logged — what ran, with which tool and model, how many attempts it took, and when it last completed.

Summary generated by qwen3.6-27b-nvidia on 2026-08-10; verification: verified.

Topics

Ranked by relevance to this paper. Hover a topic for its definition.

• drowsiness detection algorithms
• drowsiness

Information type

What kind of knowledge this paper contributes, grouped by family — independent of topic (what it is about) and method (how it was studied).

• Empirical Findings: physiological data