A Study on Physiological Data Collection and Validation in Emotion Recognition

In this paper, we recorded physiological responses to emotion-inducing stimuli in order to investigate emotion recognition based on physiological signals. Five types of physiological signals were collected: PPG, EDA, SKT, ECG, and RSP. A total of 108 healthy adult participants (52 males and 56 females, mean age 34.9 ± 11.9 years) were recruited to build the dataset. Participants were exposed to five emotion-inducing video and audio stimuli designed to evoke neutral, anger, joy, sadness, and relaxation. After viewing each stimulus, participants self-reported their emotions using categorical emotion keywords and dimensional emotion scales (arousal/valence). We validated the collected data through statistical analysis and verified the data using machine learning techniques, assessing performance through accuracy and F1 scores. In the future, we plan to develop additional emotion-inducing content beyond the current five emotions and continue data collection. The collected data will undergo a verification process and be made publicly available as a validated dataset through a web-based platform.