Investigating the Generalizability of Emotion Detection via Wearable Physiological Sensors: EmoWear Usecase

Abstract

Emotion detection is increasingly recognized as a foundational component in the evolution of intelligent eHealth systems. This paper presents a personalized methodology for identifying relevant physiological sensors for emotion recognition, using the open-access EmoWear dataset as a case study. The proposed framework applies statistical feature extraction and correlation analysis to examine the relationship between biosignals and emotional states, specifically valence, arousal, and dominance. The findings indicate that no single sensor modality universally correlates with emotional states across individuals, reinforcing the need for personalized and multimodal approaches. Notably, Electrodermal Activity (EDA) activity showed a higher correlation with valence, whereas Skin Temperature (SKT) was more closely associated with arousal; however, inter-individual variability remained significant. Overall, the analysis highlights challenges for generalizability in affective computing and emphasizes the importance of context and individual differences in emotion expression.