Πλοήγηση ανά Συγγραφέας "Froudas, Michail"

Τώρα δείχνει 1 - 1 of 1
  • Μικρογραφία εικόνας
    Τεκμήριο
    Biosignal analysis methods for the assessment of stress
    (ΕΛΜΕΠΑ, Σχολή Μηχανικών (ΣΜΗΧ), ΠΜΣ Μηχανικών Πληροφορικής, 2025-09-02) Froudas, Michail; Φρούδας, Μιχαήλ; Tsiknakis, Emmanouil; Τσικνάκης, Εμμανουήλ
    In recent years, stress has emerged as one of the most significant factors aGecting both mental and physical health. Prolonged exposure to stressful situations can lead to serious conditions such as anxiety, depression and cardiovascular disease, negatively aGecting quality of life. Early detection and accurate classification of stress is crucial for the development of eGective management and intervention strategies. In this study, we investigate the use of machine learning techniques to classify stress through physiological signals using the Wearable Stress and AGect Detection (WESAD) public dataset. WESAD includes recordings from wearable sensor devices, such as the Empatica E4 wristband and a chest belt (Respiban), and contains information from several physiological signals, such as heart rate (BVP - Blood Volume Pulse), electrodermal activity (EDA), respiratory patterns (RESP), as well as acceleration (ACC) and body temperature. Participants in the experiment were exposed to diGerent emotional states, including baseline, stress, amusement and meditation, allowing the study of the physiological response of the body to each of them. In our approach we exploit BVP and EDA signals from the Empatica E4 and develop state-ofthe-art techniques for the pre-processing of signals, the extraction of features in both the time and frequency domains, and the development of machine learning algorithms for the detection and classification of emotional states, specifically focusing on stress. Specifically, we applied Random Forest, Support Vector Machines (SVM), Logistic regression, k-nearest neighbors (KNN), Gradient boosting and Naive Bayes algorithms, and performed a comparative analysis of their performance based on accuracy. To ensure that features have similar scales standardization and normalization techniques were deployed to optimize the accuracy of the classifiers. The results of the study show that machine learning algorithms can classify stress with results achieving an accuracy of 94% using Random Forest, making stress detection systems based on wearables a reliable method for monitoring psychosomatic health. This study contributes to the development of intelligent stress recognition systems, which could be integrated into applications with the use of wearable devices, enabling early detection and tailoring personalized management strategies. Such technological developments have the potential to improve mental health care, increase productivity in the workplace and support individuals in managing their stress in a more eGective way.