Enhancing Brain-Computer Interfaces: Machine Learning Analysis of Alpha-Beta ERD and Fractal Dimension in Motor Imagery EEG

Brain-computer interfaces (BCIs) assist people with motor impairments by translating brain activity into control signals. Accurate classification of motor-related neural patterns is crucial for effective diagnosis and treatment. Event-Related Desynchronization (ERD) in the alpha and beta bands (ERD-αβ) has been widely used, while Fractal Dimension (FD), a non-linear measure of signal complexity, provides an alternative discriminative feature. This study evaluates ERD -αβ and FD using Gradient Boosting (GB), Random Forest (RF), k-Nearest Neighbours (KNN), and Support Vector Machines (SVM) on EEG data from 20 selected subjects. FD consistently outperformed ERD -αβ, with SVM achieving the highest accuracy: 86% for imagined hand movements (IHM) and 81% for imagined fist or foot movements (IFM). RF and GB also performed well with FD, while KNN showed lower accuracy. The method follows a cross-subject evaluation scheme, where models are trained and tested on data from different individuals with no data leakage. The current analysis focused on subjects with stronger EEG activation to ensure robust feature evaluation. The results suggest that FD improves classification accuracy in EEG-based BCls compared to ERD -αβ, supporting its integration into future BCI systems for improved reliability and usability.