Mental Workload Estimation Using Wireless EEG Signals

Adewale, Quadri; Panoutsos, George · 2019 · Crossref

DOI: 10.1101/755033

archive: archived pipeline: cataloged

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

Summary

This study addresses the challenge of developing fast, reliable models for estimating mental workload using electroencephalogram (EEG) signals that generalize across different tasks, subjects, and sessions. While previous research has established the correlation between EEG frequency bands and workload, existing methods often suffer from performance degradation when applied to new users or time points due to the nonstationary nature of EEG signals. Additionally, many prior studies relied on large, cumbersome electrode arrays, limiting practical real-time application. This work aims to demonstrate that a consumer-level, wireless EEG headset with a small number of electrodes can effectively support real-time cognitive monitoring. The experimental design involved eight participants (aged 19–30) performing two types of mental tasks: an n-back task (0-back for low workload, 2-back for high workload) and a mental arithmetic task (1-digit addition for low workload, 3-digit addition for high workload). EEG data were recorded using a wireless Emotiv EPOC headset with 14 electrodes. The study utilized power spectral density (PSD) as features, extracted via Welch’s method from 4-second blocks. A support vector machine (SVM) was employed for baseline classification, while Adaptive Subspace Feature Matching (ASFM), a domain adaptation technique, was applied to mitigate distribution mismatches in cross-session, cross-task, and cross-subject scenarios. Subjective workload was verified using the Rating Scale Mental Effort (RSME), and performance metrics included response time and accuracy. Results confirmed that the experimental design successfully induced distinct workload levels, with significant increases in subjective ratings, response times, and decreases in accuracy as workload increased. Within-session classification using SVM achieved high accuracies of 98.5% for the n-back task and 95.5% for the arithmetic task. ASFM significantly improved generalization capabilities: cross-session accuracies averaged 80.5% (n-back) and 74.4% (arithmetic), cross-task accuracy reached 68.6%, and cross-subject accuracies were 74.4% (n-back) and 64.1% (arithmetic). Spectral analysis revealed that alpha power decreased while theta and gamma powers increased with higher workload, consistent with prior literature. Notably, the model maintained high accuracy (96.4% for n-back) even with only 60 seconds of training data, suggesting feasibility for rapid online deployment. The significance of this research lies in demonstrating that a low-cost, wireless EEG headset can serve as a viable tool for real-time mental workload estimation in practical settings, such as driver monitoring or industrial safety. The successful application of ASFM indicates that subject- and task-independent models are achievable without extensive retraining, reducing the burden on operators and system operators. This approach offers a promising pathway for integrating passive brain-computer interfaces into adaptive systems to enhance safety and performance by facilitating task sharing between humans and machines.

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 qwen3.8-27b-gittensor on 2026-08-23; verification: pending re-verification.

Topics

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

• neuro workload indices
• workload measurement

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