We propose an objective stress assessment method based on the extraction of features from physiological time series and their classification using Support Vector Machine and K-Nearest Neighbors algorithms. For this purpose, we used an open dataset consisting of multiparametric physiological signals (electrocardiogram, electromyogram, galvanic skin response and breath signal) obtained during the execution of a driving route within the city of Boston with restful, highway and city driving periods indicative of three different stress states. To predict the driver stress level, 21 features were extracted from 122 chunks of raw signals and were subsequently managed by classification algorithms. Our analysis showed a prediction accuracy of 98.4% when all features were used, decreasing when signals from specific physiological systems were not considered. Our results highlighted that multidomain data acquisition by wearable sensors combined with appropriate classification models may represent a promising strategy to detect drivers’ stress status in an unobtrusive and objective way that can in perspective be applicable in several other fields such as in the clinics.
Fruet D., Bara C., Pernice R., Faes L., Nollo G. (2022). Assessment Of Driving Stress Through SVM And KNN Classifiers On Multi-Domain Physiological Data. In Proceedings of 2022 IEEE 21st Mediterranean Electrotechnical Conference (MELECON 2022) (pp. 920-925) [10.1109/MELECON53508.2022.9842891].
Assessment Of Driving Stress Through SVM And KNN Classifiers On Multi-Domain Physiological Data
Pernice R.;Faes L.;
2022-06-01
Abstract
We propose an objective stress assessment method based on the extraction of features from physiological time series and their classification using Support Vector Machine and K-Nearest Neighbors algorithms. For this purpose, we used an open dataset consisting of multiparametric physiological signals (electrocardiogram, electromyogram, galvanic skin response and breath signal) obtained during the execution of a driving route within the city of Boston with restful, highway and city driving periods indicative of three different stress states. To predict the driver stress level, 21 features were extracted from 122 chunks of raw signals and were subsequently managed by classification algorithms. Our analysis showed a prediction accuracy of 98.4% when all features were used, decreasing when signals from specific physiological systems were not considered. Our results highlighted that multidomain data acquisition by wearable sensors combined with appropriate classification models may represent a promising strategy to detect drivers’ stress status in an unobtrusive and objective way that can in perspective be applicable in several other fields such as in the clinics.File | Dimensione | Formato | |
---|---|---|---|
2022_Fruet_MELECON_published.pdf
Solo gestori archvio
Descrizione: Versione Editoriale pubblicata
Tipologia:
Versione Editoriale
Dimensione
868.7 kB
Formato
Adobe PDF
|
868.7 kB | Adobe PDF | Visualizza/Apri Richiedi una copia |
2022_Fruet_MELECON_postprint.pdf
accesso aperto
Descrizione: Post-print
Tipologia:
Post-print
Dimensione
635.25 kB
Formato
Adobe PDF
|
635.25 kB | Adobe PDF | Visualizza/Apri |
2022_Fruet_MELECON_preprint.pdf
accesso aperto
Descrizione: Pre-print
Tipologia:
Pre-print
Dimensione
653.38 kB
Formato
Adobe PDF
|
653.38 kB | Adobe PDF | Visualizza/Apri |
I documenti in IRIS sono protetti da copyright e tutti i diritti sono riservati, salvo diversa indicazione.