Predicting Empathy and Other Mental States During VR Sessions Using Sensor Data and Machine Learning.

Area of Science:

• Psychology and Cognitive Science
• Human-Computer Interaction
• Machine Learning Applications

Background:

• Virtual reality (VR) is recognized for its potential to foster empathy by immersing users in diverse perspectives.
• Current empathy assessment methods lack a universal standard, necessitating innovative approaches.
• Physiological responses during VR experiences offer a potential avenue for objective empathy measurement.

Purpose of the Study:

• To investigate the relationship between self-reported empathy levels and physiological responses during VR exposure.
• To develop and evaluate machine learning models for predicting state and trait empathy using physiological data.
• To introduce a novel dataset of VR videos designed to elicit empathy for research and clinical use.

Main Methods:

• 105 participants experienced 3D 360° VR videos depicting actors expressing various emotions.
• Empathy levels were assessed via self-report questionnaires.
• Physiological signals were recorded using sensors, and machine learning models (Random Forest) were employed for prediction.

Main Results:

• Random Forest models accurately predicted trait empathy (9.1% MAPE) and classified state empathy (67% balanced accuracy).
• Predictive models were developed for non-empathic arousal (78% accuracy) and distinguishing empathic vs. non-empathic arousal (79% accuracy).
• Statistical analyses explored the influence of narrative context, gender, and emotion on empathy.

Conclusions:

• Machine learning models utilizing physiological signals provide an objective and efficient method for predicting empathy levels during VR.
• This research offers a valuable dataset and predictive tools to advance empathy research and clinical applications.
• VR-based physiological monitoring shows promise as a complementary approach to traditional empathy assessment.