Classification of Emotions Based on Electrodermal Activity and Transfer Learning - a Pilot Study.

Area of Science:

• Psychophysiology
• Machine Learning
• Affective Computing

Background:

• Assessing human emotions is crucial for fields like psychology.
• Electrodermal response (EDR) is a physiological indicator of emotional arousal.
• Machine learning offers potential for objective emotion analysis.

Purpose of the Study:

• To develop and evaluate an experiment assessing emotions using EDR measurements and machine learning.
• To explore the efficacy of transfer learning and random forest classification for EDR data.
• To investigate the impact of data augmentation on classification performance.

Main Methods:

• Eliciting emotions in ten participants using film clips.
• Collecting EDR data using the Sudologger 3 device.
• Processing EDR data with continuous wavelet transformation.
• Classifying emotional states using a machine learning algorithm incorporating transfer learning and random forest classification.

Main Results:

• The experiment successfully laid a foundation for further research in emotion assessment.
• Data augmentation significantly strengthened the classification of emotional states.
• The study indicated that increased data volume would further benefit the machine learning algorithm's performance.

Conclusions:

• This pilot study demonstrates the feasibility of using EDR and machine learning for emotion assessment.
• The findings highlight the importance of data augmentation and suggest potential for larger-scale studies.
• This approach has significant implications for advancing psychological research and other related fields.