Autoencoder-Based Human Stress Detection System Using Biological Signals

Stress and depression stand as the most prevalent psychological challenges in our daily lives, significantly reducing productivity in day-to-day tasks. Stress detection model typically relies on supervised learning and physiological /behavioral markers, with promising outcomes. However, label-collec...

Celý popis

Uloženo v:
Podrobná bibliografie
Vydáno v:2024 International Conference on Recent Advances in Electrical, Electronics, Ubiquitous Communication, and Computational Intelligence (RAEEUCCI) s. 1 - 7
Hlavní autoři: Subathra, P., Malarvizhi, S.
Médium: Konferenční příspěvek
Jazyk:angličtina
Vydáno: IEEE 17.04.2024
Témata:
Agglomerative clustering
EDA
Feature extraction
Heart rate
Human factors
K-means clustering
Productivity
Psychology
Silhouette score
Stress
Supervised learning
Surveys
On-line přístup:Získat plný text
Tagy: Přidat tag
Žádné tagy, Buďte první, kdo vytvoří štítek k tomuto záznamu!
Popis
Shrnutí:Stress and depression stand as the most prevalent psychological challenges in our daily lives, significantly reducing productivity in day-to-day tasks. Stress detection model typically relies on supervised learning and physiological /behavioral markers, with promising outcomes. However, label-collecting issues, such as delay and subjective ambiguity from surveys, have not been properly addressed. This work deployed the Wearable Stress and Affect Detection (WESAD) dataset, which gathered Electrodermal Activity (EDA) and Heart Rate (HR) data from 15 subjects using wearable sensors. These physiological features are extracted using an Autoencoder and categorized using two unsupervised machine learning algorithms namely K-means and Agglomerative. Experimental results showed that agglomerative clustering obtained in this proposed method outperformed best with a silhouette score of 0.85.
DOI:10.1109/RAEEUCCI61380.2024.10547833