pith. sign in

arxiv: 2605.05120 · v1 · submitted 2026-05-06 · 💻 cs.LG · eess.SP

Physiologically Grounded Driver Behavior Classification: SHAP-Driven Elite Feature Selection and Hybrid Gradient Boosting for Multimodal Physiological Signals

Sahar Askari , Mohammad Mahdi Mirza Ali Mohammadi , Fatemeh Ensafdoust , Amin Golnari , Saeid Sanei This is my paper

Pith reviewed 2026-05-08 17:23 UTC · model grok-4.3

classification 💻 cs.LG eess.SP
keywords driving behavior classificationmultimodal physiological signalsSHAP feature selectiongradient boosting ensembleEEGEMGGSRXGBoost LightGBM
0
0 comments X

The pith

A weighted ensemble of gradient boosting models on multimodal EEG, EMG and GSR signals classifies driving behaviors at 80.91 percent accuracy.

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

The paper develops a pipeline that extracts time-domain, frequency-domain and physiological index features from synchronized EEG, EMG and GSR recordings collected while subjects drive. High dimensionality is reduced by keeping only the top 250 features ranked by SHAP values, after which Bayesian optimization tunes separate XGBoost and LightGBM classifiers whose outputs are combined by weighted soft voting. On a large labeled dataset the resulting ensemble reaches 80.91 percent test accuracy and 0.79 macro-F1, an 8 percent improvement over the strongest single-modality baseline. SHAP inspection of the retained features shows that EEG supplies the largest share of predictive weight while GSR and EMG supply necessary cues for high-arousal and motor-intensive maneuvers.

Core claim

The authors show that a weighted soft-voting ensemble of Bayesian-optimized XGBoost and LightGBM models, applied after SHAP-based retention of the top 250 features from multimodal EEG-EMG-GSR recordings, attains 80.91 percent test accuracy and 0.79 macro-F1 on driving-behavior classification. Ablation experiments establish that multimodal fusion supplies an 8 percent gain over the best single-modality model (EEG), and post-hoc SHAP analysis confirms that the selected features respect known physiological contributions of each signal modality.

What carries the argument

SHAP-based elite feature selection that retains the top 250 ranked features from multimodal physiological recordings, followed by a weighted soft-voting ensemble of Bayesian-optimized XGBoost and LightGBM models.

If this is right

  • Multimodal fusion is required because it produces an 8 percent accuracy gain over the strongest single-modality baseline.
  • The ensemble substantially outperforms both single-modality physiological models and traditional machine-learning classifiers.
  • SHAP inspection of the retained features assigns the largest predictive weight to EEG while crediting GSR and EMG with critical discrimination for high-arousal and motor-intensive maneuvers.

Where Pith is reading between the lines

These are editorial extensions of the paper, not claims the author makes directly.

  • If the feature patterns remain stable across additional driver populations, the same pipeline could be embedded in vehicle systems to flag unsafe maneuvers in real time.
  • The approach could be extended to classify related states such as fatigue or distraction by adding longitudinal recordings and retraining the ensemble on the same modality set.

Load-bearing premise

The large driving dataset contains accurately labeled behaviors and the extracted features capture physiologically stable patterns that hold across subjects and conditions.

What would settle it

Retraining and testing the identical pipeline on a fresh cohort of drivers under different conditions or labeling protocols that yields accuracy substantially below 80 percent would show that the selected features do not generalize.

Figures

Figures reproduced from arXiv: 2605.05120 by Amin Golnari, Fatemeh Ensafdoust, Mohammad Mahdi Mirza Ali Mohammadi, Saeid Sanei, Sahar Askari.

Figure 1
Figure 1. Figure 1: Representative 2-second windows of synchronized multimodal physiological signals across the four driving view at source ↗
Figure 2
Figure 2. Figure 2: Comparison of model performance across accuracy and macro-F1 metrics. The proposed ensemble view at source ↗
Figure 3
Figure 3. Figure 3: Functional performance profile comparison across the four targeted driving behaviors. The radar chart view at source ↗
Figure 4
Figure 4. Figure 4: Relative Modality Importance based on cumulative SHAP weighting. Although individual GSR and EMG view at source ↗
Figure 5
Figure 5. Figure 5: Top 25 Most Influential Physiological Features based on Mean SHAP Value. GSR-based temporal features view at source ↗
read the original abstract

An interpretable and scalable framework for decoding driving behaviors from multimodal physiological signals is proposed in this study. We utilize multimodal physiological driving behavior large-scale dataset comprising synchronized electroencephalogram (EEG), electromyography (EMG), and galvanic skin response (GSR) signals. Our approach involves rigorous preprocessing followed by a domain-specific feature extraction pipeline targeting time-domain, frequency-domain, and derived physiological indices. To address high dimensionality, we employ SHAP-based elite feature selection, retaining the top 250 features to reduce computational overhead while preserving predictive power. Hyperparameter optimization for extreme gradient boosting (XGBoost) and light gradient boosting machine (LightGBM) models is conducted using Bayesian optimization via Optuna. Finally, a weighted soft-voting ensemble is constructed to leverage the complementary strengths of both gradient boosting frameworks. The results demonstrate that the proposed ensemble achieves a test accuracy of 80.91% and a macro-F1 score of 0.79, significantly outperforming single-modality baselines and traditional machine learning models. Ablation studies confirm an 8% performance gain over the best single modality (EEG), validating the necessity of multimodal fusion. SHAP analysis further validates the physiological plausibility of the model, revealing that the EEG contributes the majority of predictive weight, GSR and EMG features provide critical discriminatory signals for high-arousal and motor-intensive maneuvers.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit. Tearing a paper down is the easy half of reading it; the pith above is the substance, this is the friction.

Referee Report

2 major / 2 minor

Summary. The paper presents an empirical framework for classifying driving behaviors from a large-scale multimodal physiological dataset (EEG, EMG, GSR). It describes preprocessing and domain-specific feature extraction (time/frequency/physiological indices), applies SHAP-based selection to retain the top 250 features, tunes XGBoost and LightGBM via Bayesian optimization (Optuna), and combines them in a weighted soft-voting ensemble. The central claims are a held-out test accuracy of 80.91% and macro-F1 of 0.79, an 8% gain over the best single modality (EEG), superiority to traditional ML baselines, and physiological plausibility of selected features via SHAP analysis.

Significance. If the performance numbers hold under subject-independent validation, the work would demonstrate the value of SHAP-driven elite feature selection and multimodal gradient-boosting ensembles for interpretable driver-behavior decoding. The explicit physiological grounding of features and the ablation showing multimodal benefit are constructive contributions to the literature on wearable/driver-monitoring systems. The empirical nature means impact hinges on reproducibility of the data partitioning and feature stability across subjects.

major comments (2)
  1. [Experimental setup / data partitioning] Experimental setup / data partitioning: the manuscript provides no description of whether the train/test split is subject-independent (e.g., leave-one-subject-out or per-driver stratified). Given documented high inter-subject variability in EEG/EMG/GSR baselines, electrode placement, and driving style, any intra-subject leakage would render the reported 80.91% accuracy, 0.79 macro-F1, and 8% multimodal gain unreliable for the stated claim of physiologically stable, generalizable patterns.
  2. [Feature-selection pipeline] Feature-selection pipeline: the choice to retain exactly the top 250 SHAP-ranked features is presented without justification or sensitivity analysis. It is unclear whether this cutoff was determined by cross-validation, elbow analysis, or domain knowledge; if critical class-discriminative features for high-arousal or motor maneuvers lie beyond rank 250, the ablation gain and SHAP physiological-plausibility conclusions would be affected.
minor comments (2)
  1. [Abstract / Methods] The abstract and methods should explicitly report the number of subjects, class distribution, and exact cross-validation scheme used for both hyperparameter tuning and final evaluation.
  2. [Ensemble construction] Notation for the weighted soft-voting ensemble (how the weights are derived and whether they are fixed or learned) is not fully specified; a short equation or pseudocode would improve clarity.

Simulated Author's Rebuttal

2 responses · 0 unresolved

We thank the referee for the constructive feedback on our manuscript. The comments highlight important aspects of experimental rigor that we will address to strengthen the presentation of our results. We respond to each major comment below and outline the corresponding revisions.

read point-by-point responses

  1. Referee: Experimental setup / data partitioning: the manuscript provides no description of whether the train/test split is subject-independent (e.g., leave-one-subject-out or per-driver stratified). Given documented high inter-subject variability in EEG/EMG/GSR baselines, electrode placement, and driving style, any intra-subject leakage would render the reported 80.91% accuracy, 0.79 macro-F1, and 8% multimodal gain unreliable for the stated claim of physiologically stable, generalizable patterns.

    Authors: We agree that the absence of an explicit description of the data partitioning strategy is a limitation in the current manuscript. The train/test split was performed in a subject-independent fashion by randomly assigning subjects to training (approximately 70% of subjects) and held-out test (30% of subjects) sets, ensuring no subject overlap. This partitioning was selected precisely to evaluate generalization across individuals given the known inter-subject variability in physiological signals. We will add a new subsection in the Methods section detailing the exact split procedure, subject counts, and confirmation of subject-independence, along with any additional validation metrics obtained under this scheme. revision: yes

  2. Referee: Feature-selection pipeline: the choice to retain exactly the top 250 SHAP-ranked features is presented without justification or sensitivity analysis. It is unclear whether this cutoff was determined by cross-validation, elbow analysis, or domain knowledge; if critical class-discriminative features for high-arousal or motor maneuvers lie beyond rank 250, the ablation gain and SHAP physiological-plausibility conclusions would be affected.

    Authors: The cutoff of 250 features was determined via preliminary cross-validation experiments in which we evaluated model performance (accuracy and macro-F1) across a range of feature counts (100, 150, 200, 250, 300, and 400) using the same SHAP ranking. Performance stabilized around 250 features with diminishing returns beyond that point, providing a practical balance between predictive power and dimensionality reduction. We acknowledge that this sensitivity analysis and the rationale were not reported in the original submission. In the revision we will include a dedicated ablation figure and table showing performance versus number of retained features, along with explicit justification for the chosen threshold. revision: yes

Circularity Check

0 steps flagged

No circularity: purely empirical ML pipeline with held-out evaluation

full rationale

The paper presents a standard supervised learning workflow: multimodal signal preprocessing, domain-specific feature extraction, SHAP-based selection of the top 250 features, Bayesian hyperparameter tuning of XGBoost and LightGBM, and weighted ensemble evaluation on a held-out test set. All reported metrics (80.91% accuracy, 0.79 macro-F1, 8% multimodal gain) are measured quantities on unseen data rather than quantities algebraically derived from fitted parameters or self-citations. No equations, uniqueness theorems, or ansatzes are invoked that would reduce the central claims to the inputs by construction. The pipeline is self-contained against external benchmarks and does not rely on load-bearing self-citations.

Axiom & Free-Parameter Ledger

2 free parameters · 2 axioms · 0 invented entities

The central claim rests on a small number of standard machine-learning assumptions plus two tunable quantities; no new physical entities are introduced.

free parameters (2)
  • elite feature count
    Fixed at 250 to balance dimensionality and performance; value is chosen rather than derived from first principles.
  • ensemble voting weights
    Soft-voting weights are set to leverage complementary strengths and are therefore fitted or validated on held-out data.
axioms (2)
  • domain assumption Multimodal physiological signals contain stable, class-discriminative information for driving behaviors
    Invoked by the feature-extraction and classification pipeline; if false, the entire performance claim collapses.
  • standard math Bayesian optimization via Optuna reliably locates competitive hyperparameters
    Standard assumption in hyperparameter search; not proven in the paper.

pith-pipeline@v0.9.0 · 5573 in / 1455 out tokens · 46914 ms · 2026-05-08T17:23:07.055061+00:00 · methodology

discussion (0)

Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.

Reference graph

Works this paper leans on

57 extracted references · 57 canonical work pages

  1. [1]

    W. H. Organization,Global status report on road safety 2023: country and territory profiles. World Health Organization, 2024

    work page 2023

  2. [2]

    Critical reasons for crashes investigated in the national motor vehicle crash causation survey,

    S. Singh, “Critical reasons for crashes investigated in the national motor vehicle crash causation survey,” tech. rep., 2015

    work page 2015

  3. [3]

    Driver crash risk factors and prevalence evaluation using naturalistic driving data,

    T. A. Dingus, F. Guo, S. Lee, J. F. Antin, M. Perez, M. Buchanan-King, and J. Hankey, “Driver crash risk factors and prevalence evaluation using naturalistic driving data,”Proceedings of the National Academy of Sciences, vol. 113, no. 10, pp. 2636–2641, 2016

    work page 2016

  4. [4]

    Differences in driving intention transitions caused by driver’s emotion evolutions,

    Y . Liu and X. Wang, “Differences in driving intention transitions caused by driver’s emotion evolutions,”Interna- tional journal of environmental research and public health, vol. 17, no. 19, p. 6962, 2020

    work page 2020

  5. [5]

    Driver’s intention recognition algorithm based on recessive markoff model,

    Z. Xing, “Driver’s intention recognition algorithm based on recessive markoff model,”Journal of Intelligent & Fuzzy Systems, vol. 38, no. 2, pp. 1603–1614, 2020

    work page 2020

  6. [6]

    Detecting driver normal and emergency lane-changing intentions with queuing network-based driver models,

    L. Bi, C. Wang, X. Yang, M. Wang, and Y . Liu, “Detecting driver normal and emergency lane-changing intentions with queuing network-based driver models,”International Journal of Human-Computer Interaction, vol. 31, no. 2, pp. 139–145, 2015

    work page 2015

  7. [7]

    Driver Gaze Region Estimation without Use of Eye Movement ,

    L. Fridman, P. Langhans, J. Lee, and B. Reimer, “ Driver Gaze Region Estimation without Use of Eye Movement ,”IEEE Intelligent Systems, vol. 31, pp. 49–56, May 2016

    work page 2016

  8. [8]

    Eye tracking in driver attention research—how gaze data interpretations influence what we learn,

    C. Ahlström, K. Kircher, M. Nyström, and B. Wolfe, “Eye tracking in driver attention research—how gaze data interpretations influence what we learn,”Frontiers in neuroergonomics, vol. 2, p. 778043, 2021

    work page 2021

  9. [9]

    Real-time monitoring of driver distraction: State-of-the-art and future insights,

    E. Michelaraki, C. Katrakazas, S. Kaiser, T. Brijs, and G. Yannis, “Real-time monitoring of driver distraction: State-of-the-art and future insights,”Accident Analysis & Prevention, vol. 192, p. 107241, 2023

    work page 2023

  10. [10]

    Eeg-based mental workload neurometric to evaluate the impact of different traffic and road conditions in real driving settings,

    G. Di Flumeri, G. Borghini, P. Aricò, N. Sciaraffa, P. Lanzi, S. Pozzi, V . Vignali, C. Lantieri, A. Bichicchi, A. Simone,et al., “Eeg-based mental workload neurometric to evaluate the impact of different traffic and road conditions in real driving settings,”Frontiers in human neuroscience, vol. 12, p. 509, 2018. 16

    work page 2018

  11. [11]

    Sanei and J

    S. Sanei and J. A. Chambers,EEG signal processing and machine learning. John Wiley & Sons, 2021

    work page 2021

  12. [12]

    Alpha-band oscillations, attention, and controlled access to stored information,

    W. Klimesch, “Alpha-band oscillations, attention, and controlled access to stored information,”Trends in cognitive sciences, vol. 16, no. 12, pp. 606–617, 2012

    work page 2012

  13. [13]

    Frontal theta as a mechanism for cognitive control,

    J. F. Cavanagh and M. J. Frank, “Frontal theta as a mechanism for cognitive control,”Trends in cognitive sciences, vol. 18, no. 8, pp. 414–421, 2014

    work page 2014

  14. [14]

    Age-related differences in pro-active driving behavior revealed by eeg measures,

    S. Getzmann, S. Arnau, M. Karthaus, J. E. Reiser, and E. Wascher, “Age-related differences in pro-active driving behavior revealed by eeg measures,”Frontiers in human neuroscience, vol. 12, p. 321, 2018

    work page 2018

  15. [15]

    Operator state in a workplace simulation modulates eye-blink related eeg activity,

    E. Alyan, E. Wascher, S. Arnau, R. Kaesemann, and J. E. Reiser, “Operator state in a workplace simulation modulates eye-blink related eeg activity,”IEEE Transactions on Neural Systems and Rehabilitation Engineering, vol. 31, pp. 1167–1179, 2023

    work page 2023

  16. [16]

    Detection of driver braking intention using eeg signals during simulated driving,

    T.-H. Nguyen and W.-Y . Chung, “Detection of driver braking intention using eeg signals during simulated driving,” Sensors, vol. 19, no. 13, p. 2863, 2019

    work page 2019

  17. [17]

    Detection of braking intention in diverse situations during simulated driving based on eeg feature combination,

    I.-H. Kim, J.-W. Kim, S. Haufe, and S.-W. Lee, “Detection of braking intention in diverse situations during simulated driving based on eeg feature combination,”Journal of neural engineering, vol. 12, no. 1, p. 016001, 2014

    work page 2014

  18. [18]

    Eeg-based driving behavior classification: Recent advances and future directions,

    W. Zhang, Y . Qin, and X. Tao, “Eeg-based driving behavior classification: Recent advances and future directions,” IEEE Transactions on Neural Systems and Rehabilitation Engineering, vol. 32, pp. 1–15, 2024

    work page 2024

  19. [19]

    Dissociating frontoparietal brain networks with neuroadaptive bayesian optimization,

    R. Lorenz, I. R. Violante, R. P. Monti, G. Montana, A. Hampshire, and R. Leech, “Dissociating frontoparietal brain networks with neuroadaptive bayesian optimization,”Nature communications, vol. 9, no. 1, p. 1227, 2018

    work page 2018

  20. [20]

    Cognitive chimera states in human brain networks,

    K. Bansal, J. O. Garcia, S. H. Tompson, T. Verstynen, J. M. Vettel, and S. F. Muldoon, “Cognitive chimera states in human brain networks,”Science advances, vol. 5, no. 4, p. eaau8535, 2019

    work page 2019

  21. [21]

    Detecting driver drowsiness based on sensors: a review,

    A. Sahayadhas, K. Sundaraj, and M. Murugappan, “Detecting driver drowsiness based on sensors: a review,” Sensors, vol. 12, no. 12, pp. 16937–16953, 2012

    work page 2012

  22. [22]

    Driving fatigue detection based on hybrid elec- troencephalography and eye tracking,

    Z. Lian, T. Xu, Z. Yuan, J. Li, N. Thakor, and H. Wang, “Driving fatigue detection based on hybrid elec- troencephalography and eye tracking,”IEEE Journal of Biomedical and Health Informatics, vol. 28, no. 11, pp. 6568–6580, 2024

    work page 2024

  23. [23]

    Comprehensive multisource learning network for cross-subject multimodal emotion recognition,

    C. Chen, Z. Li, K. I. Kou, J. Du, C. Li, H. Wang, and C.-M. V ong, “Comprehensive multisource learning network for cross-subject multimodal emotion recognition,”IEEE Transactions on Emerging Topics in Computational Intelligence, vol. 9, no. 1, pp. 365–380, 2025

    work page 2025

  24. [24]

    Visual dominance and emerging multi- modal approaches in distracted driving detection: A review of machine learning techniques,

    A. Dontoh, S. Ivey, L. Sirbaugh, A. Danyo, and A. Aboah, “Visual dominance and emerging multi- modal approaches in distracted driving detection: A review of machine learning techniques,”arXiv preprint arXiv:2505.01973, 2025

    work page arXiv 2025

  25. [25]

    A multimodal driver monitoring benchmark dataset for driver modeling in assisted driving automation,

    K. Dargahi Nobari and T. Bertram, “A multimodal driver monitoring benchmark dataset for driver modeling in assisted driving automation,”Scientific Data, vol. 11, p. 327, 2024

    work page 2024

  26. [26]

    Neurobehavioural signatures in race car driving: a case study,

    I. Rito Lima, S. Haar, L. Di Grassi, and A. A. Faisal, “Neurobehavioural signatures in race car driving: a case study,”Scientific reports, vol. 10, no. 1, p. 11537, 2020

    work page 2020

  27. [27]

    A multimodal physiological dataset for driving behaviour analysis,

    X. Tao, D. Gao, W. Zhang, T. Liu, B. Du, S. Zhang, and Y . Qin, “A multimodal physiological dataset for driving behaviour analysis,”Scientific data, vol. 11, no. 1, p. 378, 2024

    work page 2024

  28. [28]

    Xgboost: A scalable tree boosting system,

    T. Chen and C. Guestrin, “Xgboost: A scalable tree boosting system,” inProceedings of the 22nd acm sigkdd international conference on knowledge discovery and data mining, pp. 785–794, 2016

    work page 2016

  29. [29]

    Lightgbm: A highly efficient gradient boosting decision tree,

    G. Ke, Q. Meng, T. Finley, T. Wang, W. Chen, W. Ma, Q. Ye, and T.-Y . Liu, “Lightgbm: A highly efficient gradient boosting decision tree,”Advances in neural information processing systems, vol. 30, 2017

    work page 2017

  30. [30]

    Understanding driving distractions: A multimodal analysis on distraction characterization,

    M. Papakostas, K. Riani, A. B. Gasiorowski, Y . Sun, M. Abouelenien, R. Mihalcea, and M. Burzo, “Understanding driving distractions: A multimodal analysis on distraction characterization,” in26th International Conference on Intelligent User Interfaces, pp. 377–386, ACM, apr 2021

    work page 2021

  31. [31]

    Driver state recognition with physiological signals: Based on deep feature fusion and feature selection techniques,

    J. Huang, X. Huang, Y . Peng, and L. Hu, “Driver state recognition with physiological signals: Based on deep feature fusion and feature selection techniques,”Biomedical Signal Processing and Control, vol. 93, p. 106204, jul 2024

    work page 2024

  32. [32]

    Real-world driver stress recognition and diagnosis based on multimodal deep learning and fuzzy edas approaches,

    M. Amin, K. Ullah, M. Asif, H. Shah, A. Mehmood, and M. A. Khan, “Real-world driver stress recognition and diagnosis based on multimodal deep learning and fuzzy edas approaches,”Diagnostics, vol. 13, no. 11, p. 1897, 2023. 17

    work page 2023

  33. [33]

    Understanding drivers’ stress and interactions with vehicle systems through naturalistic data analysis,

    S. Milardo, P. Rathore, M. Amorim, U. Fugiglando, P. Santi, and C. Ratti, “Understanding drivers’ stress and interactions with vehicle systems through naturalistic data analysis,”IEEE Transactions on Intelligent Transportation Systems, vol. 23, pp. 14570–14581, sep 2022

    work page 2022

  34. [34]

    Drivnet: Adaptive driving behavior prediction for improving the safety of multimodal transportation in autonomous vehicle systems,

    J. R. Theivadas and S. Ponnan, “Drivnet: Adaptive driving behavior prediction for improving the safety of multimodal transportation in autonomous vehicle systems,”Multimodal Transportation, vol. 5, p. 100266, jun 2026

    work page 2026

  35. [35]

    Recognizing behavioral factors while driving: A real-world multimodal corpus to monitor the driver’s affective state,

    A. Lotz, K. Ihme, A. Charnoz, P. Maroudis, I. Dmitriev, and A. Wendemuth, “Recognizing behavioral factors while driving: A real-world multimodal corpus to monitor the driver’s affective state,” inProceedings of the Eleventh International Conference on Language Resources and Evaluation (LREC 2018), (Miyazaki, Japan), European Language Resources Association...

    work page 2018

  36. [36]

    Discrimination of cycling patterns using accelerometric data and deep learning techniques,

    A. Procházka, H. Charvátová, O. Vyšata, D. Jarchi, and S. Sanei, “Discrimination of cycling patterns using accelerometric data and deep learning techniques,”Neural Computing and Applications, vol. 33, no. 13, pp. 7603– 7613, 2021

    work page 2021

  37. [37]

    Deep multimodal learning with missing modality: A survey,

    R. Wu, H. Wang, H.-T. Chen, and G. Carneiro, “Deep multimodal learning with missing modality: A survey,” Transactions on Machine Learning Research, 2026. Survey Certification

    work page 2026

  38. [38]

    Sanei, D

    S. Sanei, D. Jarchi, and A. G. Constantinides,Body sensor networking, design and algorithms. John Wiley & Sons, 2020

    work page 2020

  39. [39]

    Detecting stress during real-world driving tasks using physiological sensors,

    J. A. Healey and R. W. Picard, “Detecting stress during real-world driving tasks using physiological sensors,” IEEE Transactions on intelligent transportation systems, vol. 6, no. 2, pp. 156–166, 2005

    work page 2005

  40. [40]

    Mmphgcn: A hypergraph convolutional network for detection of driver intention on multimodal physiological signals,

    W. Zhang, Y . Qin, and X. Tao, “Mmphgcn: A hypergraph convolutional network for detection of driver intention on multimodal physiological signals,”IEEE Signal Processing Letters, vol. 31, pp. 1730–1734, 2024

    work page 2024

  41. [41]

    Eeg-driven classification of driver mental workload in diverse environments: A dual-branch network for efficient in-vehicle applications,

    T. Liu, Y . Qin, S. Zhang, Y . Duan, and X. Tao, “Eeg-driven classification of driver mental workload in diverse environments: A dual-branch network for efficient in-vehicle applications,”IEEE Internet of Things Journal, vol. 12, pp. 34846–34862, sep 2025

    work page 2025

  42. [42]

    Phy-fusionnet: A memory-augmented transformer for multimodal emotion recognition with periodicity and contextual attention,

    T. Wu, E. Purwanto, Y . Huang, and S. Yang, “Phy-fusionnet: A memory-augmented transformer for multimodal emotion recognition with periodicity and contextual attention,”IEEE Transactions on Affective Computing, 2025

    work page 2025

  43. [43]

    Multimodal brain– computer interface for in-vehicle driver cognitive load measurement: Dataset and baselines,

    P. Angkan, B. Behinaein, Z. Mahmud, A. Bhatti, D. Rodenburg, P. Hungler, and A. Etemad, “Multimodal brain– computer interface for in-vehicle driver cognitive load measurement: Dataset and baselines,”IEEE Transactions on Intelligent Transportation Systems, vol. 25, no. 6, pp. 5949–5964, 2024

    work page 2024

  44. [44]

    Introducing wesad, a multimodal dataset for wearable stress and affect detection,

    P. Schmidt, A. Reiss, R. Duerichen, C. Marberger, and K. Van Laerhoven, “Introducing wesad, a multimodal dataset for wearable stress and affect detection,” inProceedings of the 20th ACM international conference on multimodal interaction, pp. 400–408, 2018

    work page 2018

  45. [45]

    Hatnet: Eeg-based hybrid attention transfer learning network for train driver state detection,

    S. Lin, C. Fan, D. Han, Z. Jia, Y . Peng, and S. Kwong, “Hatnet: Eeg-based hybrid attention transfer learning network for train driver state detection,”IEEE Transactions on Cybernetics, vol. 55, pp. 2437–2450, may 2025

    work page 2025

  46. [46]

    Ecg-based driver’s stress detection using deep transfer learning and fuzzy logic approaches,

    M. Amin, K. Ullah, M. Asif, A. Waheed, S. U. Haq, M. Zareei, and R. R. Biswal, “Ecg-based driver’s stress detection using deep transfer learning and fuzzy logic approaches,”IEEe Access, vol. 10, pp. 29788–29809, 2022

    work page 2022

  47. [47]

    Monitoring and analyzing driver physiological states based on automotive electronic identification and multimodal biometric recognition methods,

    S. Zhou, N. Zhang, Q. Duan, X. Liu, J. Xiao, L. Wang, and J. Yang, “Monitoring and analyzing driver physiological states based on automotive electronic identification and multimodal biometric recognition methods,”Algorithms, vol. 17, p. 547, dec 2024

    work page 2024

  48. [48]

    Cooperative identification of prolonged motor movement from eeg for bci without feedback,

    A. Falcon-Caro, J. F. Ferreira, and S. Sanei, “Cooperative identification of prolonged motor movement from eeg for bci without feedback,”IEEE Access, vol. 13, pp. 11765–11777, 2025

    work page 2025

  49. [49]

    Advances in driver behavior classification: Traditional paradigms, deep learning, generative ai, and explainable systems,

    A. Al-Rababah, S. M. Rahman, and M. Shafiullah, “Advances in driver behavior classification: Traditional paradigms, deep learning, generative ai, and explainable systems,”International Journal of Intelligent Transporta- tion Systems Research, pp. 1–39, 2026

    work page 2026

  50. [50]

    High fuel consumption driving behavior causal analysis based on lightgbm and shap,

    H. Liu, S. Chen, Y . Ma, F. Qiao, Q. Pang, Z. Zhang, and Z. Xie, “High fuel consumption driving behavior causal analysis based on lightgbm and shap,”Iranian Journal of Science and Technology, Transactions of Civil Engineering, vol. 49, no. 2, pp. 2065–2077, 2025

    work page 2065

  51. [51]

    An explainable machine learning framework for predicting driving states using electroencephalogram,

    I. Hussain, S.-J. Park, A. Azad, and S. A. Alyami, “An explainable machine learning framework for predicting driving states using electroencephalogram,”Medical Engineering & Physics, p. 104355, 2025

    work page 2025

  52. [52]

    Driver impairment detection using decision tree based feature selection and classification,

    M. Çetinkaya and T. Acarman, “Driver impairment detection using decision tree based feature selection and classification,”Results in Engineering, vol. 18, p. 101025, 2023

    work page 2023

  53. [53]

    Analyzing shap values of xgboost algorithms to understand driving features affecting take-over time from vehicle alert to driver action,

    M. Sekadakis, T. Garefalakis, P. Moertl, and G. Yannis, “Analyzing shap values of xgboost algorithms to understand driving features affecting take-over time from vehicle alert to driver action,”Displays, p. 103263, 2025. 18

    work page 2025

  54. [54]

    A multilayer stacking method base on rfe-shap feature selection strategy for recognition of driver’s mental load and emotional state,

    J. Huang, Y . Peng, and L. Hu, “A multilayer stacking method base on rfe-shap feature selection strategy for recognition of driver’s mental load and emotional state,”Expert Systems with Applications, vol. 238, p. 121729, 2024

    work page 2024

  55. [55]

    Identification and classification of driving-related stress using electrocardiogram and skin conductance signals,

    I. Marcantoni, G. Barchiesi, S. Barchiesi, C. Belbusti, C. Leoni, S. Romagnoli, A. Sbrollini, M. Morettini, and L. Burattini, “Identification and classification of driving-related stress using electrocardiogram and skin conductance signals,” in2022 IEEE International Symposium on Medical Measurements and Applications (MeMeA), pp. 1–6, IEEE, 2022

    work page 2022

  56. [56]

    Optuna: A next-generation hyperparameter optimization framework,

    T. Akiba, S. Sano, T. Yanase, T. Ohta, and M. Koyama, “Optuna: A next-generation hyperparameter optimization framework,” inProceedings of the 25th ACM SIGKDD international conference on knowledge discovery & data mining, pp. 2623–2631, 2019

    work page 2019

  57. [57]

    Driving behaviour multimodal human factors raw dataset

    X. Tao, D. Gao, W. Zhang, T. Liu, B. Du, S. Zhang,et al., “Driving behaviour multimodal human factors raw dataset.” Dataset, 2023. 19

    work page 2023