<![CDATA[This study investigates human emotion recognition using electroencephalogram (EEG) signals, focusing on the Shanghai Jiao Tong University Emotion EEG Dataset (SEED), which consists of recordings from 62 EEG channels categorized into three emotion classes: positive, neutral, and negative. The main challenges in EEG-based emotion classification include the limited amount of available data and the nonlinear, non-stationary nature of EEG signals. To address these challenges, this study evaluates the effectiveness of the Fast Fourier Transform (FFT) band power as input features and employs a stacked Long Short-Term Memory (LSTM) network as the classifier. Model validation was conducted using stratified 10-fold cross-validation, and performance was assessed using accuracy, F1-score, and Cohen’s kappa metrics. Experimental results show that the proposed method achieved an average accuracy of 89.87%, an F1-score of 90.10%, and a Cohen’s kappa value of 0.848, with minimal variation across folds, demonstrating high model stability. Unlike many recent studies that rely on image-based representations or Generative Adversarial Networks (GAN)-driven data augmentation, this study demonstrates that FFT band power combined with a sequential LSTM classifier can achieve strong performance without synthetic data generation or complex feature transformations. These findings indicate that the combination of FFT band power features and the LSTM classifier can serve as a solid baseline for further research.]]>
Copyrights © 2025
