Pith. sign in

REVIEW 3 major objections 6 minor 44 references

Estimating Markers of Driving Stress through Multimodal Physiological Monitoring

T0 review · 3 major / 6 minor · reviewed 2026-08-06 · deepseek-v4-flash

Challenge this review Re-run · record.json Download PDF Read on arXiv ↗

Pith's one-line read This paper claims that a knowledge-driven set of 12 physiological features from ECG, electrodermal, respiratory, and skin-temperature signals can detect stressor-driven driving periods at per-second resolution, reaching an AUROC of 0.812…

desk verdict Solid empirical study with a real but narrower result: the AUROC separates driving phases, not discrete stressor events, and the centered 30-s window undercuts the real-time claim. read the letter →

arxiv 2507.14146 v1 pith:ORQFVNZP submitted 2025-07-01 eess.SP

Kleanthis Avramidis , Emily Zhou , Tiantian Feng , Hossein Hamidi Shishavan , Frederico Marcolino Quintao Severgnini , Danny J. Lohan , Paul Schmalenberg , Ercan M. Dede
show 1 more author
Shrikanth Narayanan
This is my paper
classification eess.SP
keywords drivingstressmultimodalphysiologicalmonitoringelectrodermalactivityskinconductancetemperaturesimulatorgradientboostingstressordetection
verification ladder T0 review T1 audit T2 compute T3 formal

The pith

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

The reading

The paper tries to establish that short-term driving stressors leave a measurable trace in the body's autonomic signals, and that a machine-learning model built from standard physiological features can rank stressor-driven driving periods above free driving at per-second resolution. Using data from 31 participants in a driving simulator, the authors extract 12 features from ECG, electrodermal activity, respiration, and skin temperature, and train a regularized gradient-boosted tree classifier in a leave-one-subject-out scheme. The combined model reaches an AUROC of 0.812 (95% CI 0.782–0.837), with electrodermal activity the strongest single modality at 0.757. The authors also report that predicted stress correlates with slower speed, more variable steering, and condition-dependent pedal use, while most individual stressors do not produce significant short-window response slopes, suggesting that cumulative load matters alongside discrete events. A sympathetic reader would care because a physiology-only pipeline of this kind could support real-time driver state monitoring without requiring explicit event annotations.

What carries the argument

The load-bearing machinery is a knowledge-driven feature set of 12 physiological indicators computed over 30-second sliding windows with a 1-second hop: heart rate and respiratory sinus arrhythmia from ECG; mean skin conductance level, its slope, and skin conductance response frequency, amplitude, and rise time from EDA; respiratory period, depth, and volume per time from RSP; and mean temperature and temperature slope from SKT. These features are z-scored per participant using the first minute of free driving as baseline and fed to a regularized gradient-boosted tree classifier trained with leave-one-subject-out cross-validation. AUROC on held-out subjects ranks stressor-phase predictions against free-driving predictions, with bootstrap confidence intervals and permutation tests for significance.

What would settle it

Retrain the same pipeline on the same recordings with labels replaced by independently timed event annotations, or by per-event physiological response markers, and compare the AUROC on event windows against the phase-based AUROC; if the event-level AUROC drops substantially while the phase-level AUROC stays high, the claim of event-level stress estimation loses support.

Watch

Extended reading notes

Core claim

The central discovery is that a multimodal physiological feature set separates stressor-driven driving from free driving in a simulator, with the full 12-feature model reaching an AUROC of 0.812 (95% CI 0.782–0.837) while individual modalities range from 0.693 for respiration to 0.757 for electrodermal activity. The model assigns an average stress probability of 0.713 during stressor phases versus 0.266 during free driving. SHAP analysis attributes the strongest contributions to tonic skin conductance level and mean skin temperature, with lower temperature predicting stress through vasoconstriction. Behavioral analysis shows predicted stress associated with reduced average speed, increased steering variability, and condition-dependent changes in throttle and brake usage; recovery periods are heterogeneous, with only 40 percent of sessions showing decreasing stress. The paper interprets these findings to mean that stress accumulates across a drive and that model outputs track both transient events and cumulative load.

Load-bearing premise

The binary labels are defined by session phase—stressor-driven phases count as stress and free-driving phases count as not stressed—so the 0.812 AUROC assumes stress is uniformly present across whole stressor phases and absent during free driving, rather than being tied to the individual events.

Editorial extensions

If this is right

  • A physiology-only model can rank stressor periods above free-driving periods at per-second resolution, so continuous stress monitoring could run from wearable sensors without manual event logs.
  • Electrodermal activity carries the most stress signal among single modalities, while slow-moving skin temperature also discriminates; systems relying only on heart-rate features are leaving usable information unused.
  • Predicted stress ties to observable driving behavior: higher stress corresponds to slower speed, more variable steering, and condition-dependent throttle and brake patterns, offering a behavioral handle to validate a stress score.
  • Most individual stressors do not produce significant short-window response slopes, so the high AUROC largely reflects accumulated physiological state over the drive rather than clean per-event responses.

Reading between the lines

Editorial extensions of the paper, not claims the author makes directly.

  • The session-phase labels blur discrete stressors with session context; an independent event-annotation relabeling could lower the AUROC toward the true event-detection rate, and that test is ours, not the paper's.
  • If cumulative load dominates the separation, the same feature pipeline may also track fatigue or time-on-task effects, connecting this result to adjacent driver-state monitoring beyond stressor events.
  • The model's dependence on an individually z-scored baseline implies deployment needs a per-driver calibration window; the paper's reported drop with a video baseline suggests context-matched calibration is load-bearing.
Share X Bluesky LinkedIn Reddit HN

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, and a circularity audit.

Referee Report

3 major / 6 minor

Summary. The paper presents a controlled driving-simulator study with 31 participants in which ECG, EDA, respiration, and skin temperature were recorded while drivers encountered scripted stressors. Twelve knowledge-driven physiological features are extracted per second using a 30-second centered window, and an XGBoost classifier is trained under a leave-one-subject-out protocol to distinguish stressor-driven driving phases from free driving. The multimodal model attains an AUROC of 0.812 (95% CI 0.782–0.837). The paper also analyzes event-wise sensitivity, SHAP feature importance, correlations between predicted stress and vehicle-control metrics, and post-session recovery dynamics.

Significance. If the result holds, the paper provides a useful demonstration that a compact set of interpretable physiological features can rank stressor-driven driving segments above free-driving segments, and the public code and detailed protocol are valuable community assets. The subject-level cross-validation and explicit confidence intervals are strengths. However, the central claim of real-time, per-second stressor estimation is weakened by the use of a centered (non-causal) feature window and by binary labels defined at the level of entire driving phases; the paper's own event-level analyses indicate that most discrete stressors do not elicit immediate model responses. These issues must be addressed before the main claim can be accepted.

major comments (3)
  1. [Section IV-C and V-A] The binary labels are constructed from session phase: all samples during stressor-driven phases are labeled 'stress' and all free-driving samples are labeled 'not stress.' Because the free-driving block always precedes the stressor blocks within each session (Section III), the reported AUROC of 0.812 (95% CI 0.782–0.837) may reflect temporal context such as time-on-task, fatigue, or simulator novelty rather than the effect of discrete stressor events. This worry is supported by the paper's own Section V-B, which finds that most timestamped stressors do not produce significant immediate slopes in predicted stress, and by the Discussion's attribution of performance to 'cumulative stress build-up.' The permutation test in Section IV-D (random labels) does not address this alternative because it preserves the phase structure of the data. Please add an event-level evaluation that contrasts short windows around each timestamped event against matched control windows (same position within the session, no event) and report whether the model distinguishes events from time-on-task alone.
  2. [Section IV-B and VI] The feature extraction in Section IV-B uses a 30-second window centered on each time point, i.e., 15 seconds of future data. The Discussion (Section VI) claims 'real-time assessment' with 'predictions continuously updated based on the preceding 30-second window,' which is inconsistent with the centered window. Moreover, the event-sensitivity analysis in Section V-B evaluates predictions from 0 to +15 seconds after each event, so for each such prediction the centered window includes up to 30 seconds of post-event physiology; this could inflate the reported event AUROC values. Please recompute the results with a strictly causal trailing window (or explicitly present the method as non-causal and remove the real-time claims).
  3. [Section IV-D and V-A] The confidence intervals and permutation p-values are computed by bootstrap resampling individual per-second predictions from the full dataset (Section IV-D). This treats the 1 Hz samples as independent, ignoring strong temporal autocorrelation and the nested subject/session structure; consequently the 95% CIs (e.g., 0.782–0.837 for the multimodal model) are likely too narrow and the permutation test only establishes separation from random labels, not from a time-varying baseline. Please use cluster-aware bootstrapping (resample subjects or sessions) and, for the event-level claim, a permutation that shuffles event timing rather than labels.
minor comments (6)
  1. [Section IV-C] The description 'sampled N−1 participants with replacement from the training pool' is not standard leave-one-subject-out; please clarify how the 10 seeds are aggregated to produce the per-second predictions (e.g., averaging probabilities) and confirm that no participant appears in both training and testing in a given seed.
  2. [Table III] The header uses 'Pstress' while the caption refers to 'Pevents'; please reconcile the notation.
  3. [Section V-B] The AUROC construction for the 15-second event windows is unusual (comparing the within-event mean predicted probability to the entire free-driving session per participant); please describe the exact computation and whether it is per event or per participant.
  4. [Figure 5] Adding significance annotations to the left panel for the AUROC comparisons, not just the slopes, would help the reader assess event sensitivity.
  5. [Section VI] The claim that the model works 'even in the absence of explicit event annotations' is stronger than the evidence supports given the phase-based labels; please soften or qualify.
  6. [Section V-C] Consider referencing Figure 6 at the point where the SHAP analysis is introduced.

Circularity Check

0 steps flagged · score 0.0 of 10

No circularity: the AUROC and behavior correlations are held-out empirical results, not quantities forced by the model's inputs.

full rationale

I find no circular step in this paper. The central claim, an AUROC of 0.812 for distinguishing stressor-driven driving from free driving, is produced by a leave-one-subject-out evaluation protocol: predictions for each held-out subject come from a model trained on other subjects' data, and statistical significance is assessed against a permutation null distribution built from randomly permuted labels. The model output is therefore not a fitted parameter renamed as a prediction. The feature set is described as knowledge-driven from prior literature, but selecting features with known psychophysiological associations is an external, falsifiable design choice, not an input that encodes the target results. The behavior correlations are computed from trained model outputs on held-out data, not from fitting the same data used to define the claims. The only self-citation, reference 22, appears in the Background as one of several citations supporting the general fact that EDA peaks correlate with physiological arousal; it is not load-bearing for the stress-estimation result and is accompanied by independent literature. I also considered the centered 30-second window used for feature extraction versus the Discussion's characterization of a preceding 30-second window. That is a timing or validity inconsistency, not a circularity, because the model output is not equivalent by construction to any fitted constant or to the labels. There is no equation that reduces to itself, no fitted input passed off as a prediction, and no imported uniqueness theorem. The paper's result stands as an empirical, externally checkable claim that could fail on new data, which is the opposite of a derivation that is true by definition.

Assumptions & free parameters 5 free parameters · 4 assumptions · 0 invented entities

The central result depends on the psychophysiological validity of the chosen signals, the assumption that simulator events act as stress ground truth, and the choice of free driving as a neutral baseline. No new physical entities or parameters are invented; the listed free parameters are preprocessing and analysis choices that affect all reported numbers.

free parameters (5)
  • XGBoost regularization parameter = 10
    Set to 10 because of 'limited number and high variability of samples' (Section IV-C); no validation tuning but hand-chosen.
  • Feature window length = 30 seconds centered
    Chosen to balance temporal fidelity and feature reliability (Section IV-B); directly controls the real-time and causal mismatch.
  • Feature hop size = 1 second
    Increases effective training set size and enables per-second predictions (Section IV-B).
  • Nearest-neighbor imputation k = 5
    Used for missing feature values (Section IV-C); no sensitivity analysis is provided.
  • Event-response window = 15 seconds after onset
    Defines the event-level sensitivity analysis in Section V-B; results in Figure 5 depend on this window.
assumptions (4)
  • domain assumption ECG, EDA, respiration, and skin temperature features are valid markers of psychological stress during driving.
    All 12 features are drawn from prior psychophysiology literature (Sections II and IV-B); if this mapping is wrong, the model outputs do not measure stress.
  • domain assumption The timestamped simulator events induce the intended stress responses of impatience, surprise, and irritation.
    No per-event subjective or physiological ground truth is used; compliance rates and event design (Section III) are taken as evidence.
  • domain assumption The free-driving phase is a stress-free baseline for each participant.
    Individual z-scoring uses the first minute of free driving (Section IV-C), and the Limitations section reports that using the video baseline instead significantly lowers performance.
  • domain assumption The bootstrap resampling within the leave-one-subject-out protocol does not introduce subject leakage.
    The description 'sampled N-1 participants with replacement from the training pool' is ambiguous; if test subjects were not excluded, the AUC would be inflated.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Estimating Markers of Driving Stress through Multimodal Physiological Monitoring." pith.science (2026). https://pith.science/paper/ORQFVNZP

@misc{pith2026250714146,
  author       = {Pith},
  title        = {Pith review of: Estimating Markers of Driving Stress through Multimodal Physiological Monitoring},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/ORQFVNZP}},
  note         = {Machine review of arXiv:2507.14146}
}
read the original abstract

Understanding and mitigating driving stress is vital for preventing accidents and advancing both road safety and driver well-being. While vehicles are equipped with increasingly sophisticated safety systems, many limits exist in their ability to account for variable driving behaviors and environmental contexts. In this study we examine how short-term stressor events impact drivers' physiology and their behavioral responses behind the wheel. Leveraging a controlled driving simulation setup, we collected physiological signals from 31 adult participants and designed a multimodal machine learning system to estimate the presence of stressors. Our analysis explores the model sensitivity and temporal dynamics against both known and novel emotional inducers, and examines the relationship between predicted stress and observable patterns of vehicle control. Overall, this study demonstrates the potential of linking physiological signals with contextual and behavioral cues in order to improve real-time estimation of driving stress.

Figures

Figures reproduced from arXiv: 2507.14146 by the authors.

Figure 1
Figure 1. Time-series visualization of physiological signals, stress predic￾tions, and driving behavior during a sample session from the developed driving simulation. Vertical red lines indicate the onset of driving events, including environmental challenges (e.g., dense fog), vehicle interactions (e.g., slow car scenarios), abrupt braking events, up to the return to clear conditions. The plots illustrate the dynamic interpla… view at source ↗
Figure 2
Figure 2. Driving simulator used in the study. It features the cockpit with driving hardware and displays, as well as two systems for capturing biosignals and behavioral data, alongside the processing of the driving scenarios. visualization in [PITH_FULL_IMAGE:figures/full_fig_p002_2.png] view at source ↗
Figure 4
Figure 4. AUROC performance across the available training and testing modes (I: Irritation, M: Impatience, S: Surprise, All: combined). Each cell shows the average classification performance when training on the row task and testing on the column task. Higher AUROC scores and lighter colors indicate better performance. Numerical values are overlaid for clarity. Impatience (M), and Surprise (S) sessions and evaluate them acros… view at source ↗
Figures from the paper (3 more)
Figure 5
Figure 5. Figure 5: Analysis of model sensitivity to induced stressors: Left: AUROC of the multimodal classifier following each timestamped event across the three sessions. Each dot represents the subject-specific AUROC, calculated by comparing the average predicted stress probabilities w…
Figure 6
Figure 6. Figure 6: Dot plot of the four most important physiological features per category and their contribution to the stressor estimation model. Features are in descending order by contribution to the predictions, where each point represents a feature value per single participant and …
Figure 7
Figure 7. Figure 7: Characterization of stress prevalence during recovery sessions. (A-B) Distribution of individual sessions with stress scores showing a linear decrease, increase, or neither (stable) during the recovery session. Violin plots depict the overall distribution, with individ…

Discussion (0). Sign in to comment.

Reference graph

Works this paper leans on

44 extracted references · 41 canonical work pages

  1. [1]

    W. H. Organization, Global status report on road safety 2018 . World Health Organization, 2019

    work page 2018

  2. [2]

    Traffic stress, vehicular burden and well-being: a multilevel analysis,

    G. C. Gee and D. T. Takeuchi, “Traffic stress, vehicular burden and well-being: a multilevel analysis,” Social science & medicine , vol. 59, no. 2, pp. 405–414, 2004

    work page 2004

  3. [3]

    Work stress and health problems of professional drivers: a hazardous formula for their safety outcomes,

    S. A. Useche, B. Cendales, L. Montoro, and C. Esteban, “Work stress and health problems of professional drivers: a hazardous formula for their safety outcomes,” PeerJ, vol. 6, p. e6249, 2018

    work page 2018

  4. [4]

    Signal processing and machine learning for mental health research and clinical applications [perspectives],

    D. Bone, C.-C. Lee, T. Chaspari, J. Gibson, and S. Narayanan, “Signal processing and machine learning for mental health research and clinical applications [perspectives],” IEEE Signal Processing Magazine, vol. 34, no. 5, pp. 196–195, 2017

    work page 2017

  5. [5]

    Multimodal features for detection of driver stress and fatigue,

    A. N ˇemcov´a, V . Svozilov ´a, K. Bucsuh ´azy, R. Sm ´ıˇsek, M. M ´ezl, B. Hesko, M. Bel ´ak, M. Bil ´ık, P. Maxera, M. Seitl et al., “Multimodal features for detection of driver stress and fatigue,” IEEE Transactions on Intelligent Transportation Systems , vol. 22, no. 6, 2020

    work page 2020

  6. [6]

    Camera-based drowsiness reference for driver state classification under real driving conditions,

    F. Friedrichs and B. Yang, “Camera-based drowsiness reference for driver state classification under real driving conditions,” in 2010 IEEE intelligent vehicles symposium . IEEE, 2010, pp. 101–106

    work page 2010

  7. [7]

    Real-time detection of driver cognitive distraction using support vector machines,

    Y . Liang, M. L. Reyes, and J. D. Lee, “Real-time detection of driver cognitive distraction using support vector machines,” IEEE transactions on intelligent transportation systems , vol. 8, no. 2, pp. 340–350, 2007

    work page 2007

  8. [8]

    Effects of cognitive load on driving performance: The cognitive control hypothesis,

    J. Engstr ¨om, G. Markkula, T. Victor, and N. Merat, “Effects of cognitive load on driving performance: The cognitive control hypothesis,” Human factors, vol. 59, no. 5, pp. 734–764, 2017

    work page 2017

Show all 44 references
  1. [9]

    A multimodal dataset for various forms of distracted driving,

    S. Taamneh, P. Tsiamyrtzis, M. Dcosta, P. Buddharaju, A. Khatri, M. Manser, T. Ferris, R. Wunderlich, and I. Pavlidis, “A multimodal dataset for various forms of distracted driving,” Scientific data, 2017

    2017

  2. [10]

    Mea- suring neurophysiological signals in aircraft pilots and car drivers for the assessment of mental workload, fatigue and drowsiness,

    G. Borghini, L. Astolfi, G. Vecchiato, D. Mattia, and F. Babiloni, “Mea- suring neurophysiological signals in aircraft pilots and car drivers for the assessment of mental workload, fatigue and drowsiness,” Neuroscience & Biobehavioral Reviews , vol. 44, pp. 58–75, 2014

    2014

  3. [11]

    Driver fatigue detection systems: A review,

    G. Sikander and S. Anwar, “Driver fatigue detection systems: A review,” IEEE Transactions on Intelligent Transportation Systems , 2018

    2018

  4. [12]

    Daily actigraphy profiles distinguish depressive and interepisode states in bipolar disorder,

    A. Gershon, N. Ram, S. L. Johnson, A. G. Harvey, and J. M. Zeitzer, “Daily actigraphy profiles distinguish depressive and interepisode states in bipolar disorder,” Clinical psychological science , 2016

    2016

  5. [13]

    Self-supervised learning for human activity recognition using 700,000 person-days of wearable data,

    H. Yuan, S. Chan, A. P. Creagh, C. Tong, A. Acquah, D. A. Clifton, and A. Doherty, “Self-supervised learning for human activity recognition using 700,000 person-days of wearable data,” NPJ digital medicine , vol. 7, no. 1, p. 91, 2024

    2024

  6. [14]

    Sleepfm: Multi-modal representation learning for sleep across brain activity, ecg and respiratory signals,

    R. Thapa, B. He, M. R. Kjaer, H. Moore, G. Ganjoo, E. Mignot, and J. Zou, “Sleepfm: Multi-modal representation learning for sleep across brain activity, ecg and respiratory signals,” arXiv preprint arXiv:2405.17766, 2024

    2024 arXiv

  7. [15]

    Detecting stress during real-world driv- ing tasks using physiological sensors,

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

    2005

  8. [16]

    Classifying driver workload using physiological and driving perfor- mance data: two field studies,

    E. T. Solovey, M. Zec, E. A. Garcia Perez, B. Reimer, and B. Mehler, “Classifying driver workload using physiological and driving perfor- mance data: two field studies,” in Proceedings of the SIGCHI conference on human factors in computing systems , 2014, pp. 4057–4066

    2014

  9. [17]

    How the autonomic nervous system and driving style change with incremental stressing conditions during simulated driving,

    A. Lanat `a, G. Valenza, A. Greco, C. Gentili, R. Bartolozzi, F. Bucchi, F. Frendo, and E. P. Scilingo, “How the autonomic nervous system and driving style change with incremental stressing conditions during simulated driving,” IEEE Transactions on Intelligent Transportation S...

    2014

  10. [18]

    An overview of heart rate variability metrics and norms,

    F. Shaffer and J. P. Ginsberg, “An overview of heart rate variability metrics and norms,” Frontiers in public health , vol. 5, p. 258, 2017

    2017

  11. [19]

    Stress and heart rate variability: a meta-analysis and review of the literature,

    H.-G. Kim, E.-J. Cheon, D.-S. Bai, Y . H. Lee, and B.-H. Koo, “Stress and heart rate variability: a meta-analysis and review of the literature,” Psychiatry investigation, vol. 15, no. 3, p. 235, 2018

    2018

  12. [20]

    Heart rate dynamics for cognitive load estimation in a driving simulation task,

    K. R. Arutyunova, A. V . Bakhchina, D. I. Konovalov, M. Margaryan, A. V . Filimonov, and I. S. Shishalov, “Heart rate dynamics for cognitive load estimation in a driving simulation task,” Scientific Reports, 2024

    2024

  13. [21]

    Driver’s stress detection using skin potential response signals,

    A. Affanni, R. Bernardini, A. Piras, R. Rinaldo, and P. Zontone, “Driver’s stress detection using skin potential response signals,” Mea- surement, vol. 122, pp. 264–274, 2018

    2018

  14. [22]

    Multimodal estimation of change points of physiological arousal during driving,

    K. Avramidis, T. Feng, D. Bose, and S. Narayanan, “Multimodal estimation of change points of physiological arousal during driving,” in 2023 IEEE International Conference on Acoustics, Speech, and Signal Processing Workshops (ICASSPW). IEEE, 2023, pp. 1–5

    2023

  15. [23]

    Boucsein, Electrodermal activity

    W. Boucsein, Electrodermal activity. Springer, 2012

    2012

  16. [24]

    The complexity of electrodermal activity is altered in mental cognitive stressors,

    Z. Visnovcova, M. Mestanik, M. Gala, A. Mestanikova, and I. Ton- hajzerova, “The complexity of electrodermal activity is altered in mental cognitive stressors,” Computers in biology and medicine , 2016

    2016

  17. [25]

    Wearable device-based system to monitor a driver’s stress, fatigue, and drowsiness,

    M. Choi, G. Koo, M. Seo, and S. W. Kim, “Wearable device-based system to monitor a driver’s stress, fatigue, and drowsiness,”IEEE Trans. on Instrumentation and Measurement, vol. 67, no. 3, pp. 634–645, 2017

    2017

  18. [26]

    A reasoning- based framework for car driver’s stress prediction,

    G. Rigas, C. D. Katsis, P. Bougia, and D. I. Fotiadis, “A reasoning- based framework for car driver’s stress prediction,” in 2008 16th Med. Conference on Control and Automation . IEEE, 2008, pp. 627–632

    2008

  19. [27]

    Drowsiness detection based on the analysis of breathing rate obtained from real-time image recognition,

    J. Solaz, J. Laparra-Hern ´andez, D. Bande, N. Rodr ´ıguez, S. Veleff, J. Gerpe, and E. Medina, “Drowsiness detection based on the analysis of breathing rate obtained from real-time image recognition,” Transporta- tion research procedia, vol. 14, pp. 3867–3876, 2016

    2016

  20. [28]

    Breathing rate complexity features for “in-the-wild

    A. Tiwari, S. Narayanan, and T. H. Falk, “Breathing rate complexity features for “in-the-wild” stress and anxiety measurement,” in 2019 27th European Signal Processing Conference (EUSIPCO) . IEEE, 2019

    2019

  21. [29]

    Thermal infrared imaging in psychophysiology: potentialities and limits,

    S. Ioannou, V . Gallese, and A. Merla, “Thermal infrared imaging in psychophysiology: potentialities and limits,” Psychophysiology, vol. 51, no. 10, pp. 951–963, 2014

    2014

  22. [30]

    A preliminary study on driver’s stress index using a new method based on differential skin temperature measurement,

    T. Yamakoshi, K.-i. Yamakoshi, S. Tanaka, M. Nogawa, M. Shibata, Y . Sawada, P. Rolfe, and Y . Hirose, “A preliminary study on driver’s stress index using a new method based on differential skin temperature measurement,” in 2007 29th EMBC . IEEE, 2007, pp. 722–725

    2007

  23. [31]

    Development of a facial skin temperature- based methodology for non-intrusive mental workload measurement,

    C. K. Or and V . G. Duffy, “Development of a facial skin temperature- based methodology for non-intrusive mental workload measurement,” Occupational ergonomics, vol. 7, no. 2, pp. 83–94, 2007

    2007

  24. [32]

    Eeg classification of driver mental states by deep learning,

    H. Zeng, C. Yang, G. Dai, F. Qin, J. Zhang, and W. Kong, “Eeg classification of driver mental states by deep learning,” Cognitive neu- rodynamics, vol. 12, pp. 597–606, 2018

    2018

  25. [33]

    BeamNG.tech

    BeamNG GmbH, “BeamNG.tech.” [Online]. Available: https://www.beamng.tech/

  26. [34]

    NeuroKit2: A python toolbox for neurophysiological signal processing,

    D. Makowski, T. Pham, Z. J. Lau, J. C. Brammer, F. Lespinasse, H. Pham, C. Sch¨olzel, and S. H. A. Chen, “NeuroKit2: A python toolbox for neurophysiological signal processing,” Behavior Research Methods , vol. 53, no. 4, pp. 1689–1696, 2021

    2021

  27. [35]

    Statistical strategies to quantify respiratory sinus arrhythmia: are commonly used metrics equivalent?

    G. F. Lewis, S. A. Furman, M. F. McCool, and S. W. Porges, “Statistical strategies to quantify respiratory sinus arrhythmia: are commonly used metrics equivalent?” Biological psychology, vol. 89, no. 2, 2012

    2012

  28. [36]

    cvxeda: A convex optimization approach to electrodermal activity processing,

    A. Greco, G. Valenza, A. Lanata, E. P. Scilingo, and L. Citi, “cvxeda: A convex optimization approach to electrodermal activity processing,” IEEE Transactions on Biomedical Engineering , 2015

    2015

  29. [37]

    A linear model for event-related respiration responses,

    D. R. Bach, S. Gerster, A. Tzovara, and G. Castegnetti, “A linear model for event-related respiration responses,” Journal of Neuroscience Methods, vol. 270, pp. 147–155, 2016

    2016

  30. [38]

    Xgboost: A scalable tree boosting system,

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

    2016

  31. [39]

    A unified approach to interpreting model predictions,

    S. M. Lundberg and S.-I. Lee, “A unified approach to interpreting model predictions,” Advances in NeurIPS , vol. 30, 2017

    2017

  32. [40]

    Exploring the use of thermal infrared imaging in human stress re- search,

    V . Engert, A. Merla, J. A. Grant, D. Cardone, A. Tusche, and T. Singer, “Exploring the use of thermal infrared imaging in human stress re- search,” PloS one, vol. 9, no. 3, p. e90782, 2014

    2014

  33. [41]

    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 Intel- ligent Transportation Systems, vol. 23, no. 9, pp. 14 570–14 581, 2021

    2021

  34. [42]

    The proof and measurement of association between two things

    C. Spearman, “The proof and measurement of association between two things.” The American Journal of Psychology , 1961

    1961

  35. [43]

    Professional drivers’ fatigue as a problem of the modern era,

    J. Davidovi ´c, D. Pe ˇsi´c, and B. Anti ´c, “Professional drivers’ fatigue as a problem of the modern era,” Transportation research part F: traffic psychology and behaviour , vol. 55, pp. 199–209, 2018

    2018

  36. [44]

    Nasal skin temperature reveals changes in arousal levels due to time on task: An experimental thermal infrared imaging study,

    C. Diaz-Piedra, E. Gomez-Milan, and L. L. Di Stasi, “Nasal skin temperature reveals changes in arousal levels due to time on task: An experimental thermal infrared imaging study,” Applied Ergonomics, vol. 81, p. 102870, 2019

    2019

Pith tools

Reviewed August 6, 2026 · model on record in the stance chip above.