Characterisation of Cognitive Load Using Machine Learning Classifiers of Electroencephalogram Data

Wang, Qi; Smythe, Daniel; Cao, Jun; Hu, Zhilin; Proctor, Karl J.; Owens, Andrew P.; Zhao, Yifan · 2023 · Crossref

DOI: 10.3390/s23208528

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 cognitive load, which can lead to distraction and accidents, by investigating the feasibility of using electroencephalography (EEG) to classify cognitive states during driving. While EEG is widely used in cognitive research, few studies have applied it specifically to driving contexts, often relying on non-driving tasks like N-Back tests that lack ecological validity. The authors aimed to determine if machine learning classifiers could differentiate between driving conditions requiring varying levels of cognitive load using EEG data collected from a driving simulator. The methodology involved a simulator-based experiment with 20 male participants who performed four distinct driving tasks: motorway driving with no traffic, motorway driving with high traffic density (70%), urban driving with no traffic, and urban driving with moderate traffic density (30%). EEG data were recorded using a 24-channel water-based headset sampled at 256 Hz. Data preprocessing included band-pass filtering (0.2–40 Hz) and Independent Component Analysis to remove ocular artefacts. Features were extracted from four frequency bands (Theta, Alpha, Beta, and Gamma) using Power Spectrum Density and statistical measures, resulting in 384 features per data segment. These features were used to train Deep Neural Networks (DNN) and Support Vector Machine (SVM) classifiers across four classification tasks: binary high/low load for motorway, binary high/low for urban, combined high/low, and a four-class distinction of all tasks. Model performance was evaluated using 10-fold and leave-one-person-out cross-validation. The results demonstrated that DNNs significantly outperformed SVMs in all classification tasks. The best-performing model, a DNN utilizing statistical features from multiple frequency bands across all 24 channels, achieved a classification accuracy of 90.37% in the four-class task. Specifically, DNNs achieved approximately 85% accuracy in distinguishing high versus low cognitive load scenarios. The analysis revealed that Gamma and Beta frequency bands contributed more to classification accuracy than Alpha and Theta bands. Subjective ratings confirmed that tasks with traffic induced significantly higher cognitive load than those without, validating the experimental design. The significance of this work lies in its demonstration that EEG-based machine learning can effectively characterize cognitive load in realistic driving environments. By achieving high accuracy in distinguishing between different traffic densities and road types, the study supports the potential integration of EEG monitoring into Human–Machine Interfaces. This technology could enable real-time assessment of driver workload, allowing vehicles to adapt assistance levels or alert drivers to fatigue and distraction, thereby enhancing road safety.

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.

• neuro workload indices
• drowsiness detection algorithms
• workload measurement
• mental demand

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
• Theoretical Contribution: computational model