EEG-based Emotion Classification Framework Using Seagull Optimization for Feature Selection and XGBoost Hyperparameter Tuning

Human emotions play a vital role in shaping thoughts, behaviors, and decision making, yet understanding them through objective measurement continues to be a scientific challenge. It still remains one of the most complex languages that a machine can interpret. Electroencephalography(EEG) has emerged as a powerful tool throughout the years to decode emotional states from brainwave patterns. This study proposes an EEG-based emotion classification framework that integrates the Seagull Optimization Algorithm (SOA) for feature selection and XGBoost hyperparameter tuning. This research proceeded with the EEG data of 67 individuals on two contrasting stimuli recorded by the Muse 2 headset. After preprocessing and feature extraction from the key frequency bands, a four-fold cross validation approach was employed, where each fold was optimized individually using SOA for feature extraction and XGBoost hyperparameter tuning. Later, the resulting accuracies were averaged, and the most common feature masks were selected for an overall assessment of the model’s performance. The proposed framework achieved an average accuracy of 67.10 % and F1 score of 65.20% demonstrating the effectiveness of SOA-based optimization.

• Supervision