Prediction of Drowsy Driving Using EEG and Facial Expression by Machine Learning

Naito, Daichi; Hatano, Ryo; Nishiyama, Hiroyuki · 2019 · Crossref

DOI: 10.29007/v16j

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 drowsy driving, a primary cause of careless driving and traffic accidents. While existing research focuses on predicting drowsiness using specific sensors, it often neglects user preferences regarding device intrusiveness. The authors propose a flexible framework that allows users to choose between three data input methods—electroencephalogram (EEG) only, facial expressions only, or a combination of both—balancing the trade-off between prediction accuracy and ease of use. The goal is to accommodate diverse user demands, such as the desire to avoid wearing wearable devices, while maintaining effective drowsiness detection. The experimental design involved five subjects participating in a simulated driving environment using a PlayStation 2 racing game projected in a dark room to induce sleepiness. Data was collected over approximately five hours per subject using a NeuroSky MindWave Mobile single-channel dry EEG sensor and an Intel RealSense SR300 depth camera. The depth camera captured 3D coordinates of facial landmarks and head rotation angles (yaw, pitch, roll), while the EEG sensor recorded power spectral densities across eight frequency bands. Features were extracted using a sliding window approach, with windows labeled as "drowsy" or "awake" based on duration thresholds. Support Vector Machines (SVM) were employed for classification, and Random Forest was used to calculate feature importance for selection, aiming to reduce overfitting and improve model efficiency. The results demonstrated a clear hierarchy in prediction performance. Using facial expressions alone yielded the lowest accuracy (89.5%) and F-value (58.9%), though it offered the highest ease of use. EEG-only models achieved higher accuracy (97.0%) and F-value (80.3%). The combined approach of EEG and facial expressions produced the best overall performance, with an accuracy of 97.1% and an F-value of 83.0%. Feature selection analysis revealed that head rotation angles were more significant than eye size metrics for facial data. When using only the top 1% of important features, the combined model maintained high performance with a recall exceeding 90%, indicating a strong ability to detect drowsy states without missing instances. The study concludes that a multi-modal framework allows users to select a monitoring method suited to their specific needs, ranging from non-intrusive facial tracking to high-accuracy EEG monitoring. The findings confirm that combining EEG and facial data offers the most robust prediction of drowsy driving. The authors note limitations regarding the small sample size and the lack of cross-subject generalization. Future work aims to predict drowsiness levels before the onset of drowsy driving and to adapt the framework for other applications, such as smart home automation or e-learning systems.

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
• Methodological Resource: tool software