Dynamic physiological signal analysis based on Fisher kernels for emotion recognition.

Area of Science:

• Neuroscience
• Affective Computing
• Machine Learning

Background:

• Emotion recognition is crucial for physical and mental health.
• Valence/Arousal dimensions provide a framework for affective research.
• Physiological signals offer insights into emotional states.

Purpose of the Study:

• To present a novel machine learning framework for emotional regression.
• To map variable-length physiological signal sequences into a feature vector space.
• To enable accurate emotion recognition through advanced computational techniques.

Main Methods:

• Utilizing autoregressive coefficients and Hidden Markov Models (HMMs).
• Employing Fisher Kernels for signal characterization.
• Applying Support Vector Regression (SVR) on Fisher Scores for emotion recognition.
• Performing hold-out cross-validation for quantitative accuracy assessment.

Main Results:

• The proposed framework effectively performs emotional regression.
• The model demonstrates superior performance compared to static characterization methods.
• Quantitative evaluation confirms the model's accuracy in emotion recognition.

Conclusions:

• The developed machine learning framework offers a robust approach to emotion recognition.
• This method advances the analysis of emotional behavior using physiological data.
• The findings contribute to the fields of neuroscience and affective computing.