EEG-Based Emotion Classification using Nonlinear Mode Decomposition and Graph Neural Network

The neurodiagnostic application of electroencephalography (EEG) enables wide measurement of brain activity because its fast response time makes it important for research studies in addition to clinical use. This research applies the GAMEEMO dataset, which is a game-based emotional EEG dataset, to develop emotion recognition through a newly designed hybrid framework. The adopted method utilizes non-linear mode decomposition because it divides EEG signals into fundamental components to maintain important information while minimizing noise interference. The decomposed signals provide statistical and wavelet-based and frequency-domain features, which help capture multiple signal attributes. The graph neural network enables the modeling of intricate spatial relationships between different EEG channels in the data effectively. The classification performance receives further enhancement by using a graph neural network, support vector machine set-up with XGBoost. Integrated SVM with XGBoost allows their dimensional advantages in high-dimensional spaces to combine with gradient boosting strengths which result in superior decision outputs. Our method proves to be effective according to the experimental data because it achieves better results in feature extraction and classification tasks. The developed modeling system reaches an outstanding accuracy level of 99.9%, which surpasses the results of traditional methods. Together, modern signal decomposition, deep learning, and ensemble approaches greatly improve emotion identification using EEG signal, leading to more trustworthy affective computing applications.