Stress detection/classification in multimodal data

Stress detection/classification in multimodal data

Stať ve sborníku v databázi WoS či Scopus

This paper focuses on the detection and classification of stress using multimodal data. Stress monitoring is very beneficial because stress can truly negatively affect the quality of life of an individual. Chronic stress may lead to various health issues, including cardiovascular, autoimmune, and mental diseases, and in severe cases, it can result in premature death. The WAUC database, which includes data from mental stress, physical stress, and a combination of both, was used for this work. It contains data from 48 subjects, of which 26 subjects were used. For this study, electrocardiogram, galvanic skin response, respiration, and temperature signals were used. A variety of machine learning models were trained for several classification tasks. The best model for classifying stress into six groups is the Support Vector Machines (SVM) classifier, with an F1 score of 82.5% for the training dataset and 41.2% for the testing dataset. The SVM classifier shows the best results when stress is classified into three groups representing different levels of physical stress with an F1 score of 73.8% for the testing dataset. The Boosted Trees classifier shows the best results when physical stress is detected with an F1 score of 97.5% for the testing dataset. The best model for stress detection, regardless of whether it is physical or mental, is the SVM with an F1 score of 75.9% for the testing dataset. © 2025, Brno University of Technology. All rights reserved.

This paper focuses on the detection and classification of stress using multimodal data. Stress monitoring is very beneficial because stress can truly negatively affect the quality of life of an individual. Chronic stress may lead to various health issues, including cardiovascular, autoimmune, and mental diseases, and in severe cases, it can result in premature death. The WAUC database, which includes data from mental stress, physical stress, and a combination of both, was used for this work. It contains data from 48 subjects, of which 26 subjects were used. For this study, electrocardiogram, galvanic skin response, respiration, and temperature signals were used. A variety of machine learning models were trained for several classification tasks. The best model for classifying stress into six groups is the Support Vector Machines (SVM) classifier, with an F1 score of 82.5% for the training dataset and 41.2% for the testing dataset. The SVM classifier shows the best results when stress is classified into three groups representing different levels of physical stress with an F1 score of 73.8% for the testing dataset. The Boosted Trees classifier shows the best results when physical stress is detected with an F1 score of 97.5% for the testing dataset. The best model for stress detection, regardless of whether it is physical or mental, is the SVM with an F1 score of 75.9% for the testing dataset. © 2025, Brno University of Technology. All rights reserved.

biological signals; classification; detection; machine learning; multimodal data; stress

biological signals; classification; detection; machine learning; multimodal data; stress

JORDÁNOVÁ, N.; NĚMCOVÁ, A.

2026

29.04.2025

Brno University of Technology

Brno

978-80-214-6321-9

Proceedings II of the Conference Student EEICT

23

26

4

https://www.eeict.cz/eeict_download/archiv/sborniky/EEICT_2025_sbornik_1.pdf