PCA-Based Feature Optimization for Motor Imagery EEG Classification Using Machine Learning Models

Abstract

A fundamental component of Brain–Computer Interface (BCI) systems is Motor imagery (MI)-based electroencephalogram (EEG) signal categorization. However, inter-subject variability and high dimensionality frequently restrict its performance. This paper proposes a thorough machine learning (ML) framework for categorization using a recently collected dataset that hasn’t been thoroughly examined in previous research. In order to improve MI-EEG signal recognition, this paper introduces a hybrid ML approach that combines feature standardization, dimensionality reduction using Principal Component Analysis (PCA), and multi-model classification. To improve discriminative power and cut down on redundancy, the collected features are first normalized and condensed to a compact representation using PCA. A stratified train–test split is then used to train and assess several supervised learning algorithms, including Naïve Bayes (NB), Support Vector Machine (SVM), Random Forest (RF), Logistic Regression (LR), K-Nearest Neighbors (KNN), Gradient Boosting (GB), Light Gradient Boosting Machine (LightGBM), and Extreme Gradient Boosting (XGBoost). According to experimental data, NB performs the lowest (52.70%), whereas Logistic Regression has the best average classification accuracy of 85.91%, followed by SVM (79.16%) and RF (77.68%). Ensemble and boosting techniques like LightGBM (75.37%), and XGBoost (75.06%) yield competitive results, demonstrating their subject-matter resilience. The results show that PCA greatly improves classification efficiency without sacrificing high accuracy. All things considered, the suggested architecture provides a dependable and scalable MI-EEG classification solution and has potential for BCI applications in real time.