Assessing EEG Responses to Stress in Virtual Reality Scenarios

Stress is a complex and multifactorial emotion that significantly affects the psychophysical well-being of individuals by disrupting the homeostasis of the human body. The present study aims to assess EEG responses to stress and stressors in Virtual Reality scenarios. VR has been chosen for its ability to create controlled, immersive environments that can elicit realistic responses from participants. Two immersive VR environments were developed: one designed to induce stress and the other to provide a mitigated, less stressful experience. Each virtual environment is divided into seven parts, with tasks related to specific stressors (lack of time, lack of knowledge, lack of control, lack of salvation, no or too many alternatives, lack of engagement, and lack of self-confidence) selected based on previous literature. The experiment involved 61 participants who interacted with both VR environments and performed tasks related to the identified stressors. EEG signal, Skin Conductance and pulse-wave analysis (PPG) were recorded during the sessions using an EEG headset (Emotiv EPOC Flex) and the Shimmer3 GSR. Following the VR experience, participants completed a Self-Assessment Manikin (SAM) scale questionnaire to evaluate their emotional state, thereby validating the VR environments. This thesis focuses on the EEG signal analysis. EEG data were analyzed to derive the power of different frequency bands, which were then used to calculate stress indicators such as valence, arousal, dominance, stress, and frontal alpha asymmetry (FAA). The questionnaire responses served as ground truth for labelling the indicators extracted from the EEG signals, facilitating the training of four machine learning classifiers: KNN, RF, SVM, and XGBoost. Three different class labelling methods were employed, all demonstrating high accuracy (80\% to 100\%) in class recognition. The emotional response was evident in brainwave activity, with higher beta and gamma wave activation during the stressful experience and higher alpha and theta activation during the less stressful experience. Statistical analysis using t-tests confirmed significant differences in physiological responses between the two conditions. The results showed strong agreement between questionnaire responses and EEG indicators and validate the EEG as a mean to assess stress condition.