An Explainable Machine Learning Approach Based on Statistical Indexes and SVM for Stress Detection in Automobile...

Area of Science:

• * Driver safety and automotive engineering.
• * Biomedical signal processing and machine learning applications.

Background:

• * Increasing car accidents are linked to driver stress, necessitating effective stress detection methods.
• * Electromyographical signals (EMG) offer insights into physiological stress responses.
• * Statistical Time Features (STFs) can capture subtle signal changes indicative of stress.

Purpose of the Study:

• * To investigate the efficacy of STFs in detecting driver stress using EMG signals.
• * To evaluate different machine learning classifiers and kernels for stress event detection.
• * To explore model explainability for enhanced comprehension of algorithm performance.

Main Methods:

• * Analysis of electromyographical signals from drivers.
• * Extraction and application of various statistical time features (e.g., variance, standard deviation).
• * Implementation and comparison of Support Vector Machine (SVM) classifiers with different kernels (e.g., cubic).

Main Results:

• * Variance and standard deviation were identified as highly effective STFs for stress detection.
• * A Support Vector Machine classifier with a cubic kernel achieved an Area Under the Curve (AUC) of 0.97.
• * Different SVM kernels demonstrated varying efficacy when trained with STFs, impacting model selection.

Conclusions:

• * STFs, particularly variance and standard deviation, are effective for detecting driver stress events.
• * SVM with a cubic kernel provides high accuracy in identifying stress using EMG-derived features.
• * Model explainability is crucial for understanding algorithm performance and selecting appropriate machine learning models for driver stress detection.