An End-to-End Brain Computer Interface System for Mental Workload Estimation through Hybrid Deep Learning Model

Sharma, Vipul; Ahirwal, Mitul Kumar · 2024 · Crossref

DOI: 10.1007/s44230-024-00086-y

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 accurately estimating mental workload (MWL) to enhance safety and efficiency in brain-computer interface (BCI) systems and high-stress professions. The authors argue that existing methods often rely on subjective assessments or handcrafted feature extraction from electroencephalogram (EEG) signals, which can be biased or inefficient. To overcome these limitations, the study proposes an end-to-end deep learning framework that automatically extracts spatial and temporal features from raw EEG data without manual engineering. The research specifically targets binary classification (low vs. high workload) and ternary classification (low, moderate, and high workload) using the Simultaneous Task EEG Workload (STEW) dataset. The methodology employs a hybrid deep learning model consisting of cascaded one-dimensional convolutional neural networks (1DCNN) and bidirectional long short-term memory (BLSTM) layers. The 1DCNN layers extract spatial features from multivariate time-series EEG data, while the BLSTM layers capture temporal dependencies in both forward and backward directions to mitigate the vanishing gradient problem. The dataset comprises EEG recordings from 45 subjects performing simultaneous audio-visual tasks, sampled at 128 Hz across 14 channels. Data augmentation was performed using overlapping windows of 512 samples with a shift of 128 samples, resulting in 6,615 samples. The model was trained using stochastic gradient descent with a custom learning rate schedule and evaluated using holdout validation, fivefold, and sevenfold cross-validation. The results demonstrate that the proposed 1DCNN-BLSTM architecture achieves high classification accuracy, surpassing previous state-of-the-art studies on the STEW dataset. For binary classification, the model achieved an accuracy of 97.89% on the holdout test set and 96.77% with sevenfold cross-validation. For ternary classification, the model attained 95.87% accuracy on the holdout set and 95.36% with sevenfold cross-validation. The study reports detailed performance metrics, including precision, recall, and F1 scores, confirming the model's robustness despite occasional instability in loss curves during specific cross-validation folds. The confusion matrices indicate effective separation of workload classes, with minor misclassifications primarily attributed to class imbalance in the dataset. The significance of this work lies in its demonstration that end-to-end deep learning can effectively replace manual feature extraction for MWL assessment, offering a more objective and automated solution. By achieving superior accuracy with raw EEG signals, the proposed framework provides a viable tool for real-time monitoring of mental states in safety-critical environments. The study contributes a novel hybrid architecture and a flexible framework applicable to both binary and ternary classification tasks, addressing the growing need for reliable mental health and workload monitoring systems in the face of rising mental stress-related issues.

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

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