Bimodal affect recognition based on autoregressive hidden Markov models from physiological signals.

Area of Science:

• Affective computing
• Physiological signal processing
• Machine learning for emotion recognition

Background:

• Accurately determining a person's emotional state during communication is a significant challenge.
• Physiological biosignals, including electrocardiogram (ECG) and electrodermal activity (EDA), reflect emotional states via the peripheral nervous system.
• Computational emotion recognition from physiological data remains an open research problem.

Purpose of the Study:

• To develop and evaluate a method for accurate recognition of six distinct emotions.
• To leverage ECG and EDA signals for computational emotion analysis.
• To address the challenge of real-time emotion detection in human-computer interaction.

Main Methods:

• Utilized autoregressive hidden Markov models (AR-HMMs) to model electrodermal activity (EDA) signals.
• Applied heart rate variability (HRV) analysis to electrocardiogram (ECG) and EDA data.
• Employed a hierarchical classification approach, first distinguishing positive/negative valence, then specific emotions.

Main Results:

• Achieved an average accuracy of 88.6% in distinguishing between six emotions (happiness, sadness, surprise, fear, anger, disgust).
• Validated the method on a new dataset comprising 30 participants.
• Demonstrated effective emotion recognition with limited training samples using Linear Discriminant Analysis (LDA).

Conclusions:

• The proposed method effectively recognizes six emotions from ECG and EDA signals with high accuracy.
• The integration of AR-HMMs for EDA and LDA for classification offers a robust approach to emotion recognition.
• The hierarchical classification strategy enhances accuracy and reduces the need for extensive training data.