Unveiling Human Emotions: Analysing Emotional Contagion through Electrodermal Activity via Machine Learning Techniques

Emotions are often conceptualized as a complex and structured family, organized within a meaningful hierarchy. Regardless, basic emotions are considered innate, meaning that they are inherent to individuals, and thus, in any social interaction, there will inevitably be some exposure to emotion. It is an instinctive tendency for humans to synchronize with the emotional states they perceive during interactions. Evidence suggests that emotions can indeed be contagious, leading to both mental and physical arousal (Schachter and Singer, 1962). Beginning with the seminal work of William James and Carl Lange in 1884, the periphery-feedback theory of emotion established a foundational understanding of the correlation between physiological changes and emotional shifts. However, it left unresolved questions regarding the mechanisms translating stimuli events into peripheral physiological changes and the subsequent internal link to emotional arousal. This thesis critically examines the utilization of Electrodermal Activity (EDA) measurement research to unravel emotional responses. While EDA's role in regulating internal temperatures is acknowledged, its robust correlation with emotional arousal is repeatedly affirmed. Emotional arousal, influenced by environmental stimuli, prompts fluctuations in EDA, reflecting shifts in eccrine sweat gland activity. Notably, both positive and negative stimuli evoke increased arousal, leading to heightened skin conductance, indicating the broad sensitivity of EDA to various emotional dimensions. While the study of emotional contagion regarding basic emotions from a psychological perspective has notable literature support, the scientific exploration of this phenomenon in less common emotional states, such as social and pure contagious emotions, is less established. The proposed work aims to study the EDA signal by exclusively utilizing skin conductance without incorporating other biological signals. This study addresses this gap by analysing the results of an experimentation conducted to explore emotional contagion from different perspectives. In particular, the experiment involved distinct phases with different groups of subjects: in the first phase, the impact of truly eliciting online videos on subjects viewing them for the first time is examined; the second phase focuses on how other groups of subjects react to expressions extracted from the videos of the first phase's subjects. Attention is given to social emotions (embarrassment, disgust, and schadenfreude) and contagious emotions (laughter, yawning, and mirror pain), comparing them differently to express varied emotional conditions. The analysis of the EDA signal was conducted with Ledalab, a MATLAB toolbox. Different Machine Learning models (Decision Tree, Support Vector Machine (SVM), k-Nearest Neighbour (kNN) and Naive Bayes) were explored to classify emotions based on their arousal levels for the subjects. Furthermore, the Shapiro-Wilk test was used to assess the normality of the data distributions. Since the data were not normally distributed, the Mann-Whitney U test was employed to identify significant differences between Phase 1 and Phase 2. Additionally, the False Discovery Rate (FDR) adjustment was applied to account for multiple comparisons, ensuring accurate interpretation of the results. Despite a limited dataset, this work aims to thoroughly anal the underexplored and vast realm of emotions, with the future aspiration of leveraging these findings to enhance applications in fields such as human