Driver Cognitive Load Classification Based on Physiological Data—Case Study 7

He, Dengbo; Risteska, Martina; Donmez, Birsen; Chen, Kaiyang · 2021 · Intelligent Computing for Interactive System Design

DOI: 10.1145/3447404.3447428

archive: archived pipeline: cataloged verified

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

Summary

This paper addresses the challenge of estimating driver cognitive load in real-time to enhance vehicle safety, particularly regarding distraction from auditory-verbal secondary tasks. As in-vehicle infotainment systems become more prevalent, drivers face increased cognitive demands that can impair visual scanning and driving performance. The authors argue that intelligent vehicle systems must detect not only the presence of distraction but also the level of cognitive load to adapt interfaces appropriately, such as by filtering notifications or delaying non-critical information. While various measures exist for cognitive load assessment, physiological signals like Electroencephalogram (EEG), heart rate (HR), and galvanic skin response (GSR) are identified as valuable for real-time detection due to their sensitivity to cognitive shifts. The study presents a case study using data from a driving simulator experiment involving 33 participants. Each participant completed three drives under different cognitive load conditions: no secondary task, a lower difficulty cognitive task (1-back), and a higher difficulty cognitive task (2-back). Physiological data were collected using a consumer-grade wireless EEG headband (Muse), along with ECG and GSR sensors. The EEG signals were processed to extract power features from five frequency bands (delta, theta, alpha, beta, and gamma) at two electrode positions, while HR and GSR were calculated as single features. To mitigate individual differences, the data were normalized against each participant’s baseline no-task responses. Six supervised machine learning models—k-nearest neighbors (kNN), artificial neural network (ANN), support vector machine (SVM), Naïve Bayes (NB), decision tree, and linear discriminant analysis (LDA)—were trained to classify the three cognitive states. The results demonstrated that normalizing physiological data significantly improved classification accuracy, boosting the best model’s performance by 26.5%. The SVM model achieved the highest classification accuracy of 79.4% for distinguishing between no-task, 1-back, and 2-back conditions, followed by ANN and kNN at 75.8%. Confusion matrices indicated that models were most accurate in identifying the no-task condition and the high-load 2-back condition, with moderate accuracy for the intermediate 1-back task. The authors note that while earlier studies using research-grade EEG achieved higher accuracies (85–96%), those studies typically addressed simpler binary classification problems (task vs. no-task) rather than distinguishing between multiple levels of cognitive load. The significance of this work lies in demonstrating that affordable, consumer-grade physiological sensors can effectively classify nuanced levels of driver cognitive load. The findings suggest that real-time adaptive systems can distinguish between low and high taskload, allowing for targeted interventions that avoid nuisance alarms during low-load periods. The study highlights the importance of data normalization to account for individual variability and suggests that future research should explore models capable of capturing temporal state transitions, such as Hidden Markov Models or Recurrent Neural Networks, to further improve detection accuracy.

Key finding

Support Vector Machines achieved the highest classification accuracy of 79.4% in distinguishing between no-task, low-difficulty, and high-difficulty cognitive load conditions using normalized physiological data from consumer-grade sensors.

Methodology

simulator

Sample size: 33

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.

StageOutcomeToolModelPromptAttemptsCompleted
discover success 1 2026-05-28
archive success canonical_url 1 2026-06-06
extract success cached 3 2026-06-10
clean success clean 1 2026-06-07
chunk success chunk 1 2026-06-07
embed success embed Qwen/Qwen3-Embedding-8B 1 2026-06-07
enrich success openalex 5 2026-07-02
promote success 1 2026-06-04
summarize success llm qwen3.6-27b-prismaquant summ-v5 2 2026-06-10
tag success vector_similarity 15 2026-06-11
verify success 2 2026-06-10

Summary generated by qwen3.6-27b-prismaquant on 2026-06-10; verification: verified.

Topics

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

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).