# A Feasibility Study of Using an In-Ear EEG System for a Quantitative Assessment of Stress and Mental Workload

## TL;DR

This study explores a new in-ear EEG device for measuring stress and mental workload, showing promising results for real-time monitoring.

## Contribution

A novel in-ear EEG system is introduced and validated for stress and mental workload assessment with practical usability.

## Key findings

- Cross-subject stress classification achieved 77% accuracy, and within-subject stress regression had an average R2 of 0.76 ± 0.20.
- Mental workload classification reached 70-80% accuracy for arithmetic and finger tapping tasks.
- Alpha and beta band features were most important for model performance, while mental rotation showed lower accuracy.

## Abstract

While electroencephalography (EEG) is effective for assessing stress and mental workload, its widespread adoption is currently hindered by the complex setup of most existing EEG systems. This article presents a new in-ear EEG system and investigates its feasibility for developing robust models to quantify stress and mental workload levels. The system consists of a single-channel EEG acquisition device that has a similar form factor as user-generic earpieces. All electrodes including passive, reference and bias electrodes were put on the ear, which optimized the device’s usability. We validated the system through two experiments with 66 subjects to collect EEG data under varying stress and mental workload conditions. We developed classification and regression models to predict stress and mental workload levels from the data. Cross-subject stress classification achieved 77% accuracy, while within-subject stress regression yielded an average R2 of 0.76 ± 0.20. Two-class mental workload level classification reached accuracies between 70% and 80% for the arithmetic and finger tapping tasks. Feature importance analysis revealed that frequency-domain EEG features, particularly in the alpha and beta bands, significantly contributed to the models’ performance. However, we observed lower within-subject feature variation and model accuracy for the mental rotation, potentially due to the distance between brain regions engaged and the device’s recording site. Our findings demonstrate the potential of using the presented EEG device to monitor stress and mental workload in real-time.