Advancing Cognitive Load Detection in Simulated Driving Scenarios Through Deep Learning and fNIRS Data
DOI: 10.20944/preprints202506.2162.v1
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 challenge of driver cognitive overload in the era of conditionally automated driving. As Advanced Driving Assistance Systems (ADAS) shift responsibility between human and machine, maintaining situational awareness is vital to prevent accidents caused by distraction or inattention. The research aims to develop a robust method for real-time cognitive load detection using Functional Near-Infrared Spectroscopy (fNIRS) and deep learning, specifically within a high-fidelity simulated driving environment. The motivation stems from the need to move beyond traditional statistical methods, which often lack generalizability, toward data-driven models capable of capturing complex, non-linear physiological patterns across diverse populations. The experimental design involved 38 healthy participants with valid driver’s licenses who performed a dual-task paradigm in a driving simulator. The simulator replicated a Toyota Fortuner SUV and utilized motion platforms and panoramic displays to enhance realism. Participants drove under challenging environmental conditions—nighttime and heavy rain—while simultaneously performing an auditory-modified n-back task to induce varying levels of cognitive workload (0-back, 1-back, and 2-back). Prefrontal cortex hemodynamic activity was recorded using a high-density fNIRS system. The raw data were pre-processed and segmented into temporal windows of 10, 20, and 30 seconds using both overlapping and non-overlapping strategies. These segments were classified using EEGNet, a compact convolutional neural network originally designed for EEG data, trained with the Adam optimizer over 200 epochs. The results demonstrated that classification performance was significantly influenced by hyperparameter tuning and temporal segmentation strategies. A learning rate of 0.001 yielded the highest accuracy. Specifically, the model achieved 100% classification accuracy when using overlapping windows and 97% accuracy with non-overlapping windows. These findings indicate that the sliding window approach effectively captures short-term fluctuations in brain activation, enhancing the model's ability to distinguish between different cognitive load levels. The study confirms that fNIRS signals, particularly changes in blood oxygenation in the prefrontal cortex, correlate with increased mental effort during multitasking driving scenarios. The significance of this work lies in its demonstration of the viability of combining fNIRS with deep learning for scalable, real-time cognitive load monitoring. By achieving near-perfect accuracy in a controlled yet realistic setting, the study highlights the potential for integrating such systems into future driver-assistance technologies. The findings underscore the importance of temporal modeling in physiological signal analysis, suggesting that deep learning models can generalize better than traditional methods. This approach offers a pathway for developing adaptive systems that can detect driver distraction or overload and intervene to enhance road safety, particularly in complex driving conditions where human factors are prone to error.
Key finding
The EEGNet deep learning model achieved 100% accuracy in classifying cognitive load levels from fNIRS data when using overlapping temporal windows and a learning rate of 0.001.
Methodology
simulator
Sample size: 38
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. Discovered via author_sweep_intake on 2026-05-28.
| Stage | Outcome | Tool | Model | Prompt | Attempts | Completed |
|---|---|---|---|---|---|---|
| discover | success | author_sweep | — | — | 2 | 2026-05-28 |
| archive | success | openalex | — | — | 9 | 2026-06-06 |
| extract | success | cached | — | — | 3 | 2026-06-10 |
| clean | success | clean | — | — | 1 | 2026-06-04 |
| chunk | success | chunk | — | — | 1 | 2026-06-04 |
| embed | success | embed | Qwen/Qwen3-Embedding-8B | — | 1 | 2026-06-04 |
| enrich | success | — | — | — | 1 | 2026-05-28 |
| 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).
- Empirical Findings: physiological data
- Methodological Resource: tool software
- Theoretical Contribution: computational model