Predicting Driver Fatigue in Automated Driving with Explainability

Zhou, Feng; Alsaid, Areen; Blommer, Mike; Curry, Reates; Swaminathan, Radhakrishnan; Kochhar, Dev; Talamonti, Walter; Tijerina, Louis · 2021 · arXiv

archive: archived pipeline: cataloged verified

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

Summary

This study addresses the critical safety challenge of driver fatigue in automated driving, where monotony and underload can lead to drowsiness and impaired takeover performance. While machine learning models have been widely used for fatigue detection, their "black-box" nature limits trust and the ability to derive actionable intervention strategies. To bridge this gap, the authors propose an explainable framework combining eXtreme Gradient Boosting (XGBoost) for prediction and SHapley Additive exPlanations (SHAP) for interpretation. The research aims to accurately predict fatigue using non-invasive physiological and behavioral measures while uncovering the specific relationships between these inputs and fatigue levels. The experimental design involved 20 participants in a high-fidelity driving simulator configured with SAE Level 3 automation. To elicit passive fatigue, drivers engaged the automated system for approximately 60 minutes without secondary tasks or hazardous events. The ground truth for fatigue was PERCLOS (percentage of eyelid closure over the pupil), measured via eye-tracking goggles. The model utilized 11 predictor variables, including heart rate, breathing rate, ECG, steering torque, and posture, collected via wearable sensors and vehicle interfaces. The XGBoost regression model was trained on 58,846 one-second samples, and SHAP was employed to quantify feature importance globally and explain individual predictions locally. The XGBoost model significantly outperformed six other machine learning algorithms, achieving a root-mean-squared error (RMSE) of 3.847, a mean absolute error (MAE) of 1.768, and an adjusted R² of 0.996. SHAP analysis revealed that a subset of five physiological features—average heart rate, heart rate variability, average breathing rate, and their respective standard deviations—provided optimal performance, rendering behavioral measures like steering angle less critical. The study identified specific non-linear relationships: average heart rate and breathing rate were generally negatively correlated with fatigue, while heart rate variability exhibited a V-shaped relationship with PERCLOS. These findings suggest that low arousal (indicated by low heart and breathing rates) and specific variability patterns are strong indicators of fatigue in automated contexts. The significance of this work lies in its dual contribution to accuracy and interpretability. By identifying that physiological measures are superior predictors in automated driving, the study supports the development of minimally invasive monitoring systems. Furthermore, the explainable insights offer concrete guidelines for countermeasures; for instance, interventions such as high-tempo music or haptic breathing guidance could be tailored to modulate heart and breathing rates to maintain optimal arousal. This approach moves beyond mere detection, providing domain knowledge that can inform the design of active safety systems to mitigate fatigue during the critical transition from automated to manual control.

Key finding

XGBoost + SHAP yields highly accurate (R^2 0.996) and interpretable fatigue prediction in automated driving, enabling targeted intervention during takeover.

Methodology

other

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 discover_arxiv on 2026-05-04 (4 acquisition events logged).

Summary generated by qwen3.6-27b-prismaquant on 2026-06-10; verification: verified_with_issues.

Topics

Ranked by relevance to this paper. Hover a topic for its definition.

• drowsiness detection algorithms
• truck driver fatigue

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
• Theoretical Contribution: computational model, theory or model