Reward-Penalty Weighted Ensemble for Emotion State Classification from Multi-Modal Data Streams.

Area of Science:

• Affective Computing
• Machine Learning
• Signal Processing

Background:

• Single-modal data streams limit emotion classification accuracy.
• Multi-modal data streams are essential for improving real-time emotion recognition.
• Existing online ensemble methods use fixed parameters, often determined by trial and error, impacting performance.

Purpose of the Study:

• To introduce a novel Reward-Penalty-based Weighted Ensemble (RPWE) for real-time multi-modal emotion classification.
• To address the limitations of fixed parameters in traditional online ensemble approaches.
• To enhance the accuracy and robustness of emotion classification using multi-modal physiological data.

Main Methods:

• Development of the Reward-Penalty-based Weighted Ensemble (RPWE) algorithm.
• Utilizing multi-modal physiological data streams for emotion classification.
• Testing RPWE on benchmark datasets (DEAP and AMIGOS).
• Comparison of RPWE against existing online ensemble methods.

Main Results:

• The first experiment validated the effectiveness of RPWE with base stream classifiers for real-time emotion classification.
• The second experiment demonstrated RPWE's superiority over popular online ensemble approaches.
• RPWE achieved strong performance in terms of average balanced accuracy and F1-score.

Conclusions:

• RPWE offers a robust and effective solution for real-time multi-modal emotion classification.
• The proposed method significantly improves upon existing online ensemble techniques.
• RPWE shows promise for applications requiring accurate and immediate emotion recognition from physiological signals.