CPO-CLSA: A Hybrid Emotion Classification and Detection Method Using Enhanced Feature Extraction from Wearable Physiological Signals

To overcome limitations in feature extraction and improve classification accuracy in emotion models, a novel CPOCLSA model using wearable physiological signals is proposed. The CLSA (CNN-LSTM Spatial Attention) architecture features a dual-branch design, combining CNN for spatial feature extraction, LSTM for temporal dependencies in emotional transitions, and a spatial attention mechanism for efficient feature selection. The Crested Porcupine Optimizer (CPO) algorithm is used for automatic hyperparameter tuning, enhancing training efficiency and generalization. Experiments on Non-EEG, WESAD, and NEEG-Sub datasets show the CPO-CLSA model achieves 99.8% accuracy on Non-EEG and 99% on WESAD, demonstrating its effectiveness in dynamic adaptability and precise emotion classification. This model offers a robust solution for real-time emotion recognition in wearable devices, with significant potential for mental health monitoring and adaptive human-computer interaction systems.