EEG-Based Brain–Computer Interface for Intelligent Wheelchair Control: A Machine Learning Approach

Abstract

Electroencephalography EEG -based motor imagery brain–computer interfaces BCIs offer a noninvasive means of control for individuals with severe motor impairments, but their translation into physically actuated systems remains challenged by the inherent uncertainty of EEG decoding. This thesis presents the design, implementation, and evaluation of an EEG-based motor imagery BCI for wheelchair control, integrating a Filter Bank Common Spatial Pattern FBCSP feature extraction pipeline with a comparative evaluation of three classifiers, Support Vector Machine SVM, k-Nearest Neighbors KNN, and Linear Discriminant Analysis LDA, on the BCI Competition IV Dataset 2a under a subject-dependent paradigm. Using five-fold stratified cross-validation, SVM achieved the highest mean accuracy 90.7 ± 2.6% and Cohen’s Kappa 0.876 ± 0.035, and was subsequently evaluated on an independent, unseen dataset via a replay based Hardware-in-the-Loop HIL methodology, achieving 85.9% accuracy. A confidence-based command validation stage further improved reliability, raising accuracy among accepted predictions to 97.6% at 60.7% coverage. The classifier output was interfaced with a physical differential-drive wheelchair prototype governed by an encoder-based closed-loop steering control scheme. A multi-tiered safety strategy, comprising confidence gating, self-terminating steering, transition braking, communication timeout supervision, and a hardware emergency cutoff, was incorporated to constrain the consequences of residual classification uncertainty. These results demonstrate that careful integration of classification, command validation, and embedded safety design can yield a robust and practical EEG-based wheelchair control framework, despite the imperfect reliability of motor imagery decoding.