Evaluation of time series changes of mental workload during automobile driving with application of anomaly detection

CHIHARA, Takanori; SAKAMOTO, Jiro · 2021 · Crossref

DOI: 10.5100/jje.57.2c4-1

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 challenge of evaluating time-series changes in mental workload (MWL) during automobile driving, a critical factor in preventing inattentive driving caused by cognitive overload. While previous research by the authors proposed quantifying overall MWL using eye movement parameters and One-Class Support Vector Machine (OCSVM) anomaly detection, this work specifically investigates how cognitive load fluctuations, induced by secondary tasks, affect MWL dynamics over time. The motivation stems from the understanding that MWL is not constant during driving and that real-time monitoring could enhance safety. The experimental design involved 12 healthy students (ages 20–24) with driver’s licenses performing a driving task on a simulator (UC-win/Road) along a 1 km urban course. Participants completed three laps: the first and third laps involved driving only, while the second lap included an N-back secondary task to manipulate cognitive load. Two conditions were tested for the secondary task: 1-back (lower load) and 3-back (higher load). Eye and head movements were recorded using an image sensor. Four eye movement parameters—standard deviation of gaze angle, standard deviation of eye rotation angle, head movement contribution ratio, and blink rate per minute—were extracted using a 90-second sliding window. These parameters served as features for the OCSVM model, which was trained on the first 50% of the first lap’s data (no secondary task) using scikit-learn with an RBF kernel ($\gamma=0.001$, $\nu=0.05$). The remaining data served as test sets. The results demonstrated that the proposed method could visualize time-series changes in MWL. The anomaly score, representing estimated MWL, was significantly higher during the 3-back condition compared to other intervals. Furthermore, the proportion of data classified as "abnormal" (indicating high MWL) was significantly lower in the training data (first lap, no task) than in the 1-back and 3-back conditions. The median proportion of abnormal data for both 1-back and 3-back conditions was approximately 0.9, confirming that the N-back task increased MWL detectable by the model. However, the study noted that the proportion of abnormal data in the first lap (before the secondary task) was also relatively high (~0.54), suggesting that a threshold of zero for anomaly detection leads to excessive false positives. The study concludes that OCSVM-based anomaly detection using eye movement parameters can effectively visualize MWL increases and decreases corresponding to the loading and unloading of secondary tasks. It confirms the method's ability to quantitatively distinguish between different levels of MWL (e.g., 3-back vs. 1-back). However, the findings highlight the necessity of optimizing the anomaly detection threshold to reduce false positives for accurate real-time MWL evaluation. This work contributes to the development of non-invasive, real-time monitoring systems for driver cognitive state assessment.

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.

• 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
• Methodological Resource: metric or index
• Theoretical Contribution: theory or model