Deep Temporal Modeling and Ensemble Fusion for Multimodal Emotion Recognition from Physiological Signals

Short summary

Researchers tested deep learning models (LSTM, TCN, Transformer) for emotion recognition using wearable sensor signals from the WESAD dataset. Transformer models achieved the highest accuracy, while an ensemble strategy combining all three models reached 98.91% accuracy with sensor fusion. The results confirm that multi-sensor fusion and ensemble methods are effective for building robust physiological emotion-recognition systems.

• •Transformer models outperform LSTM and TCN in multimodal emotion recognition
• •Ensemble method combining all three architectures achieves 98.91% accuracy
• •Sensor fusion with wearable physiological signals enables robust emotion detection

Generated with AI, which can make mistakes.