Predicting Reduced Driver Alertness on Monotonous Highways

Larue, Gregoire S.; Rakotonirainy, Andry; Pettitt, Anthony N. · 2015 · Crossref

DOI: 10.1109/mprv.2015.38

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 safety risk posed by reduced driver alertness on monotonous highways, where low cognitive stimulation leads to inattention and increased crash likelihood. While electroencephalography (EEG) provides accurate alertness measurements, it is impractical for in-vehicle use due to cost, discomfort, and interference. The research aims to predict alertness decrements using surrogate measures from feasible in-vehicle sensors, enabling real-time monitoring and early warning systems. The experiment utilized a driving simulator with 25 participants who completed four 40-minute sessions under varying road monotony conditions. Driver alertness was objectively measured via EEG, serving as the ground truth reference. Surrogate data were collected using an eye tracker (blink frequency, eye closure), electrocardiogram (heart rate), electrodermal activity (skin conductance), and vehicle dynamics (lane position, steering). The researchers applied Principal Component Analysis to reduce data redundancy and tested five machine learning models: Generalized Linear Mixed Models, Bayesian Networks (Hidden Markov Models), Neural Networks, Support Vector Machines, and Classification and Regression Trees. Models were trained individually for each participant to account for inter-individual differences. Results indicated that linear models (GLMMs) and Bayesian networks failed to accurately predict low alertness states, largely due to violated assumptions and the complex, non-linear nature of alertness fluctuations. Support Vector Machines and Decision Trees showed moderate performance but were insufficient for reliable prediction. Neural Networks emerged as the superior model, achieving 90% accuracy in detecting reduced alertness episodes, with a 10% false positive and false negative rate, and an Area Under the Curve of 0.91. Crucially, the Neural Network model could predict alertness decrements up to five minutes in advance with over 85% accuracy, using inputs such as blink frequency, skin conductance, and time to line crossing. The findings demonstrate that machine learning, specifically Neural Networks, can effectively predict driver alertness lapses on monotonous roads using practical surrogate sensors. This capability allows for early intervention before driving performance significantly degrades. The study highlights the necessity of personalized models due to high inter-individual variability in driving behavior and physiological responses. These results support the development of intelligent in-vehicle systems that monitor driver state in real-time, offering a viable countermeasure against crashes caused by hypovigilance on highways.

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.

• drowsiness detection algorithms
• vigilance
• drowsiness
• telematics crash prediction
• truck driver fatigue
• situational awareness

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
• Theoretical Contribution: computational model