Emotion Detection with EEG and Peripheral Physiological Data Using Enhanced ID Convolutional LSTM Networks

Abstract

This research focuses on classifying human emotions using a hybrid 1D Convolutional Long Short-Term Memory (CNN-LSTM) neural network with emotion recognition, a critical aspect of affective computing, benefits significantly from integrating Electroencephalogram (EEG) signals and peripheral physiological data. This study proposes a novel hybrid deep learning framework combining enhanced identity-preserving (ID) mechanisms with Convolutional Long Short-Term Memory (Conv LSTM) networks. Leveraging Transformer-based modules and AI-driven denoising algorithms, the proposed model enhances EEG signal quality, ensures real-time edge deployment, and integrates multi-domain features (time-series, frequency, and spatial) with peripheral signals such as galvanic skin response (GSR) and heart rate variability (HRV). To improve transparency and trust, SHAP (Shapley Additive explanations) is employed for model explainability. EG and peripheral physiological data. The study utilizes the DEAP dataset, comprising 32 EEG channels and 8 peripheral physiological channels. The proposed 1D CNN-LSTM model achieved 91.19% accuracy for valence and 91.51% for arousal, outperforming traditional classifiers such as Support Vector Machines (SVM), K-Nearest Neighbours (KNN), and Random Forest (RF). The study also investigates emotion classification performance based on different brain lobes and hemispheres, revealing that the frontal lobe and left frontal region combined with peripheral data deliver the highest accuracy. Experimental validation on multimodal datasets, including DEAP and AMIGOS, demonstrates that the framework achieves robust emotion classification accuracies exceeding 95%, outperforming traditional methods. Applications include mental health monitoring, human-computer interaction (HCI), and adaptive learning systems, highlighting its transformative potential in real-world settings.