Cross-Subject Motor Imagery EEG Classification using CSP and Machine Learning

This study investigates motor imagery EEG classification using a machine learning pipeline based on the PhysioNet EEG Motor Movement/Imagery dataset. The proposed approach includes signal preprocessing, band-pass filtering (8–30 Hz), epoch extraction, and spatial feature extraction using Common Spatial Patterns (CSP). Multiple classification models were evaluated, including Support Vector Machine (SVM), Logistic Regression, Linear Discriminant Analysis (LDA), and Random Forest. Experiments were conducted under both within-subject and cross-subject (LOSO) evaluation protocols. Results show that CSP significantly improves classification performance, achieving up to 92.86% accuracy in within-subject evaluation. However, performance drops under cross-subject conditions, with a best average accuracy of 51.11% using SVM. These findings highlight the challenges of generalization in EEG-based brain–computer interfaces.