Multivariate Multiscale Entropy: An Approach to Estimating Vigilance of Driver

Ahammed, Kawser; Ahmed, Mosabber Uddin · 2021 · Crossref

DOI: 10.21203/rs.3.rs-30557/v2

archive: archived pipeline: cataloged

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

Summary

This paper addresses the gap in driver vigilance estimation techniques that fail to analyze cognitive states within the complexity domain. While existing methods rely on time-domain or frequency-domain features, the authors propose applying Multivariate Multiscale Entropy (MMSE) to characterize vigilance. The study aims to demonstrate how MMSE can distinguish between awake, tired, and drowsy states using electroencephalogram (EEG) and electrooculogram (EOG) signals, and to validate this approach using machine learning. The research utilizes a multimodal dataset from the SEED-VIG database, which includes EEG signals from posterior and temporal sites and forehead EOG signals, recorded at 1000 Hz and downsampled to 200 Hz. The signals were preprocessed with a 1–75 Hz band-pass filter. Differential entropy (DE) features were extracted from five standard frequency bands (delta, theta, alpha, beta, gamma) and from a higher-resolution 2 Hz frequency grid to capture detailed vigilance dynamics. The MMSE method was applied to these DE features and to Percentage of Eye Closure (PERCLOS) values. To validate the distinguishing ability of MMSE, the mean multivariate sample entropy values across scales for the three cognitive states were used as input features for a Support Vector Machine (SVM) classifier. The SVM classification was evaluated using a 5-fold cross-validation approach. The results indicate that MMSE analysis reveals statistically significant differences (p < 0.01) in complexity among brain EEG, forehead EEG, and EOG signals. Specifically, forehead EEG signals exhibited higher complexity curves compared to brain EEG and EOG signals. The mean multivariate sample entropy values across all scales were significantly different among the three cognitive states: awake (1.0828 ± 0.4664), tired (0.7841 ± 0.3183), and drowsy (0.2938 ± 0.1664). The SVM classifier, using these entropy features, achieved a classification accuracy of 76.2% in distinguishing the cognitive states. The confusion matrix showed that the drowsy state was identified with 100% sensitivity and 85.7% specificity, while the awake state had 57.1% sensitivity and 92.9% specificity. The significance of this work lies in demonstrating that complexity-based features derived from MMSE can effectively characterize driver vigilance, offering a complementary approach to traditional time- and frequency-domain methods. The findings suggest that forehead EEG signals are particularly informative in the complexity domain. Although the accuracy of 76.2% is slightly lower than some conventional methods cited in the literature (83.6%–88.6%), the study provides a novel framework for building programmable vigilance detection systems. The authors conclude that MMSE can be practically implemented for continuous attention monitoring, though they note that the current dataset context (monotonous highway driving) may limit generalizability to high-mental-load scenarios like city driving. Future work should explore real-world applications and neurofeedback integration.

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.

• drowsiness detection algorithms
• drowsiness
• microsleep
• neuro workload indices
• vigilance
• distraction 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
• Methodological Resource: validation psychometrics, tool software