A Stacked GRU-LSTM Framework for Improved Emotion Detection in EEG Data with Feature Optimization

Emotion is a complex psychological state that impacts human behavior, cognition, and physiology, often triggered by internal or external stimuli.Accurate recognition of emotions is crucial for understanding human experiences and has significant applications in various fields, including human-computer interaction, mental health, and affective computing.Electroencephalography (EEG) has emerged as a valuable tool for emotion recognition due to its ability to non-invasively measure brain activity and detect patterns associated with different emotional states.However, capturing complex emotions from EEG data remains challenging due to inherent noise and signal complexity, which can limit the performance of existing methods.This study presents an emotion recognition system based on a deep learning (DL) model utilizing a stacked gated recurrent unit (GRU) and long short-term memory (LSTM) network, enhanced by feature optimization through L1 norm regularization.The model was designed to categorize positive, neutral, and negative emotions from EEG data by capturing both long-term and short-term dependencies inherent in EEG signals.The EEG dataset used for this study was obtained from the publicly available Kaggle EEG Brainwave Dataset: Feeling Emotions.This dataset consists of EEG recordings associated with different emotional states, providing a robust basis for training and validating the proposed model.Feature optimization using L1 norm regularization refined the feature set, reducing computational complexity and mitigating overfitting by focusing on the most relevant features.The model attained 95.55% accuracy, demonstrating its effectiveness in detecting and categorizing emotions from EEG signals.