Emilija Kizhevska and Mitja Luštrek

Abstract
Empathy is a multifaceted concept with both cognitive and
emotional components that plays a crucial role in social
interactions, prosocial behavior, and mental health. In our
study, empathy and general arousal were induced via VR,
with physiological signals measured and ground truth collected
through questionnaires. Data from over 100 participants were
collected and analyzed using multiple machine learning models
and classification algorithms to predict empathy based on
physiological responses. We explored different data balancing
techniques and labeled data in multiple ways to enhance
model performance. Our results show that they are effective in
detecting general arousal, empathy, and differentiating between
non-empathic and empathic arousal, but the models encountered
difficulties with precise emotion detection. The dataset extracted
at 5-second intervals and models using Random Forest and
Extreme Gradient Boosting showed the best performance. Future
work will focus on refining emotion detection through advanced
modeling techniques and investigating gender differences in
empathy.