Automatic stress detection in car drivers based on non-invasive physiological signals using machine learning techniques

Siam, Ali I.; Gamel, Samah A.; Talaat, Fatma M. · 2023 · openalex

DOI: 10.1007/s00521-023-08428-w

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 critical issue of driver stress, a significant contributor to traffic accidents and fatalities, by proposing an automatic stress detection technique (SDT) for automotive drivers. While many existing studies rely on laboratory-controlled environments or vehicle motion metrics that are susceptible to external factors, this research focuses on non-invasive physiological signals—specifically electrocardiogram (ECG), electromyogram (EMG), galvanic skin response (GSR) from both hands and feet, and respiration rate. These signals are chosen for their direct correlation with autonomic nervous system activity and their reliability regardless of lighting or weather conditions. The proposed system is designed to integrate with Driver Assistance Systems (DAS) to monitor mental states and potentially trigger interventions to enhance safety. The methodology employs a three-phase pipeline: biosignal pre-processing, feature extraction, and classification. Using the `drivedb` dataset, which contains recordings from multiple drivers across various driving conditions (rest, city, and highway), the authors segmented signals into one-minute partitions. Pre-processing involved filtering to remove noise, such as baseline wander in ECG signals and DC components in GSR signals. From each partition, ten statistical features were extracted, including peak counts, interval means, and root mean square values. These features were standardized and split into training (70%) and testing (30%) sets. Six machine learning classifiers—K-Nearest Neighbors (KNN), Support Vector Machine (SVM), Decision Tree (DT), Logistic Regression (LR), Random Forest (RF), and Multilayer Perceptron (MLP)—were trained using grid-search optimized hyperparameters to distinguish between stressed and relaxed states. The experimental results demonstrate that the Random Forest (RF) classifier significantly outperformed the other models. The RF model achieved a classification accuracy of 98.2%, with a sensitivity of 97% and a specificity of 100%. In comparison, the SVM achieved 95.2% accuracy, while KNN and LR both reached 91.02%. The RF model also maintained a high precision and F1-score of 0.98. The study further analyzed feature importance, indicating that the selected physiological features effectively discriminated between stress levels. The robust performance of the RF classifier, combined with its relatively short training time, highlights its suitability for real-time implementation in vehicle safety systems. The significance of this work lies in its validation of a high-accuracy, non-invasive method for detecting driver stress using readily accessible physiological sensors. By achieving near-perfect specificity and high accuracy, the proposed SDT offers a reliable tool for early stress identification, which can mitigate the risk of accidents caused by impaired judgment or compromised performance. The findings support the integration of AI-driven physiological monitoring into modern Driver Assistance Systems, providing a proactive approach to managing driver mental states and improving overall road safety.

Key finding

The Random Forest classifier achieved the highest performance in detecting driver stress from physiological signals, yielding a classification accuracy of 98.2%, sensitivity of 97%, and specificity of 100%.

Methodology

dataset

Sample size: 10

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 scout_discovery on 2026-05-08.

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
• stress driving
• 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
• Theoretical Contribution: computational model