Decoding emotions and unveiling stress: a non-invasive approach through sequential feature extraction and multiclass classifiers

• Published in: Health and technology 2024-11, Vol.14 (6), p.1149-1160
• Main Authors: Upadhaya, Shraddha, Brahma, Biswajit, K.S, Hareesha, Panigrahi, Ranjit, Bhoi, Akash Kumar
• Format: Article
• Language: English
• Subjects: Accuracy; Algorithms; Biological and Medical Physics; Biomedical Engineering and Bioengineering; Biomedicine; Biophysics; Classification; Computational Biology/Bioinformatics; Computer vision; Datasets; Emotions; Engineering; Feature extraction; Heart diseases; Heart rate; Hypertension; Machine learning; Medicine/Public Health; Mental health; Original Paper; Physiology; R & D/Technology Policy; Regression analysis; Regression models; Respiration; Sensors; Skin; Stress; Support vector machines; Time series; Wearable computers
• ISSN: 2190-7188; 2190-7196
• DOI: 10.1007/s12553-024-00900-4

Purpose Stress is widespread in the modern world. It is a complex fusion of psychological and physiological tension that leads to various health issues, such as heart disease, high blood pressure, and widespread anxiety. Although monitoring emotions, especially stress, is critically challenging, however, to tackle this challenge head-on, advancements in machine learning have paved the way for unraveling the complexities of human emotions and detecting early signs of stress. Methods In this exploratory study, we introduce an innovative framework built on a Sequential Feature Extractor (SFE), which collaborates seamlessly with k-Nearest Neighbor (KNN), linear Support Vector Classifier (SVC), Support Vector Machine (SVM), and Logistic Regression (LR). The model identifies seven crucial features in this context through refined preprocessing methods. Results The SFE + KNN model stands out by leveraging its attributes, displaying remarkable precision and an F1-Score of 88.00% when detecting stress. Furthermore, concerning individual emotions, this model excels in various ways. The SFE + SVM methodology accurately identifies Transient emotions at a rate of 94.00% and flags Baseline emotions with a perfect score of 100.00%. Amusement is deftly grasped with 79.00% accuracy using SFE + LR. Meanwhile, the SFE + SVC approach astutely recognizes Stress at 84.00% and Meditation at 92.00%. These results underscore the model’s capability to untangle the complex tapestry of human sentiments and stress responses successfully. Conclusions The study utilizes the publicly available WESAD Dataset and achieves impressive accuracy levels in detecting stress and various emotions. The approach taken in this study contributes to understanding human emotional experiences and coping mechanisms, leading to improved resilience and emotional intelligence.