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

In three linesEvaluation of deep learning models (LSTM, TCN, Transformer) on WESAD dataset for emotion recognition from physiological signals (wrist/chest sensors). Late-fusion ensemble achieves 98.91% accuracy and 98.56% macro-F1. Transformer excels in multimodal, TCN in wrist-only settings.

Summary generated by Claude — human-verified