Emotion Recognition from Physiological Signals Collected with a Wrist Device and Emotional Recall

Area of Science:

• Affective computing
• Human-computer interaction
• Physiological computing

Background:

• Effective emotion recognition using physiological signals is crucial for real-world affective engineering applications.
• A gap exists in translating laboratory-based emotion recognition findings to higher technology readiness levels.
• This research investigates the feasibility of emotion recognition outside controlled laboratory settings.

Purpose of the Study:

• To assess the feasibility of emotion recognition using physiological measurements in a minimally-invasive, real-world-like setup.
• To explore the impact of subject dependency on emotion recognition accuracy.
• To evaluate machine learning models for classifying emotions based on physiological data.

Main Methods:

• Utilized an Empatica wristband to collect physiological data (blood volume pressure, electrodermal activity, skin temperature, acceleration).
• Employed autobiographical emotion memory tasks for emotional recall.
• Applied feature-based supervised learning, specifically Support Vector Machines, with various segmentation methods.
• Conducted 10-fold cross-validation and leave-one-subject-out cross-validation.

Main Results:

• Subject-dependent models achieved up to 75% accuracy for classifying four emotions.
• Subject-independent models (leave-one-subject-out) yielded low accuracy, around 32%.
• Initial subject-independent 10-fold cross-validation showed accuracy barely above random guessing (36%).

Conclusions:

• Emotion recognition from physiological signals is highly subject-dependent.
• Subject-specific data significantly enhances classification accuracy.
• Further research is needed to overcome subject-dependency challenges for robust real-world emotion recognition systems.