Optimized node-level capsule graph neural network for subject-independent emotion recognition from EEG signals

Area of Science:

• Neuroscience
• Machine Learning
• Signal Processing

Background:

• Subject-independent emotion detection from Electroencephalography (EEG) is challenging due to limited labeled datasets.
• Existing models often struggle with generalization across diverse populations and emotional states.
• Accurate emotion recognition is crucial for various applications, including mental health monitoring and human-computer interaction.

Purpose of the Study:

• To develop a robust subject-independent emotion detection system using EEG.
• To enhance the performance of a Node-Level Capsule Graph Neural Network (NCGNN) for emotion recognition.
• To improve the accuracy, precision, and recall of emotion classification from EEG data.

Main Methods:

• Utilized Vibrational Mode Decomposition (VMD) for feature extraction from EEG signals.
• Developed a Node-Level Capsule Graph Neural Network (NCGNN) for emotion classification.
• Integrated the Piranha Foraging Optimization Algorithm (PFOA) to optimize NCGNN parameters for enhanced performance.
• Implemented the proposed NLCGNN-SIER-EEG model in Python and evaluated its performance using metrics like Accuracy, Precision, Recall, F1 score, and ROC.

Main Results:

• The proposed NLCGNN-SIER-EEG technique demonstrated significant improvements over existing methods.
• Achieved higher accuracy, precision, and recall compared to SIER-EEG-VMD-DL, ERS-TLE-DCNN, and EEH-HER-ANN.
• Specifically, accuracy was improved by 19.57%, 24.37%, and 34.15%; precision by 22.12%, 26.82%, and 28.52%; and recall by 23.26%, 28.17%, and 29.43%, respectively.

Conclusions:

• The Piranha Foraging Optimization Algorithm effectively enhances the Node-Level Capsule Graph Neural Network for subject-independent emotion detection from EEG.
• The proposed NLCGNN-SIER-EEG model offers a more accurate and robust solution for real-world emotion recognition applications.
• This approach addresses the limitations of small or biased datasets in EEG-based emotion detection.