Comparative Analysis of Feature Selection Methods using PCA, Mutual Information, and RFE for EEG Eye State Classification

Abstract

Electroencephalogram (EEG)-based eye state classification is an important task in Brain-Computer Interface (BCI) systems and real-time monitoring applications such as fatigue detection and assistive technologies. This study aims to evaluate the effectiveness of feature selection methods, including Principal Component Analysis (PCA), Mutual Information (MI), and Recursive Feature Elimination (RFE), in improving classification performance for EEG-based eye state detection. The EEG Eye State dataset obtained from Kaggle, consisting of 14,980 instances and 14 features, is used as the experimental benchmark. The proposed methodology involves data preprocessing, feature selection, and classification using Support Vector Machine (SVM). The performance of each method is evaluated using accuracy, precision, recall, F1-score, and confusion matrix. The results show that the baseline model without feature selection achieves the highest performance, with an accuracy of 0.6939 and an F1-score of 0.5399. While RFE achieves the highest precision (0.9068), it suffers from low recall, and PCA yields the lowest overall performance due to its variance-based transformation approach. The findings indicate that feature selection does not necessarily improve classification performance in EEG datasets with moderate dimensionality. Instead, preserving the full feature set allows better representation of inter-channel dependencies, leading to improved classification results. This study provides important insights into the trade-offs between dimensionality reduction and classification performance, contributing to the development of efficient and reliable EEG-based systems.