Enhancing EEG-Based Limbs Movement Classification through Advanced Machine Learning Techniques for Motor Imagery

Motor Imagery (MI) mental process is a pivotal component of Brain-Computer Interface (BCI) systems, facilitating interaction with external environments for individuals with motor impairments. This study aims to evaluate motor imagery classification beyond hand movements, incorporating foot tasks, using simpler Machine Learning (ML) models (XGBoost, LightGBM, RandomForest) with a small number of Electroencephalogram (EEG) channels (8-11) and without data augmentation. Specifically, we propose the MIRA system, capable of performing rehabilitation tasks. The experimental results show that the adopted models perform well on the correct hand/left-hand classification, achieving accuracy, recall, precision, and F1 scores of approximately 73% for both classes. However, the dataset is unbalanced in the hand/foot task, resulting in precision, recall, and F1 values for the foot class ranging from 15% to 41 %, despite an average accuracy of 77%. The proposed system opens new avenues for research and practical applications to improve the quality of life of individuals with motor impairment.