An ensemble deep learning framework for emotion recognition through wearable devices multi-modal physiological...

Area of Science:

• Computer Science
• Affective Computing
• Human-Computer Interaction

Background:

• Wearable fitness trackers enable health monitoring.
• Emotion recognition via wearables can enhance human-computer interaction.
• Accurate emotion detection is crucial for developing emotionally aware systems.

Purpose of the Study:

• To propose and experimentally analyze an emotion recognition system using wearable devices.
• To evaluate both discrete and dimensional emotion models.
• To compare the performance of different wearable devices for emotion recognition.

Main Methods:

• An ensemble deep learning architecture combining Long Short-Term Memory (LSTM) and Gated Recurrent Unit (GRU) models was employed.
• The EMOGNITION database, featuring physiological data from Samsung Galaxy Watch, Empatica E4, and MUSE 2 EEG headband, was utilized.
• Nine discrete emotions and the 2D Valence-Arousal dimensional model were investigated using various bio-signal combinations.

Main Results:

• High classification accuracies were achieved: 99.14% for Samsung Galaxy Watch and 99.41% for MUSE 2 for discrete emotions.
• For the Valence-Arousal model, Samsung Galaxy Watch achieved 97.81% accuracy for Valence and 72.94% for Arousal.
• The study demonstrated promising results for emotion recognition using wearable technology compared to existing methods.

Conclusions:

• Wearable devices, particularly the Samsung Galaxy Watch and MUSE 2 EEG, show significant potential for accurate emotion recognition.
• The proposed deep learning approach effectively captures temporal dependencies in physiological signals for emotion detection.
• This research contributes to the advancement of emotionally aware systems and human-computer interaction.