Feature extraction feature selection and dimensionality reduction techniques for brain computer interface

Brain–Computer Interfaces (BCIs) enable direct communication between the brain and computers, offering non-muscular control for applications such as rehabilitation, performance augmentation, and target recognition. Modern non-invasive BCIs rely on EEG signals and machine learning techniques for interpretation, but challenges of robustness, real-time processing, and nonstationarity persist. This thesis investigates feature manipulation—encompassing extraction, selection, and dimensionality reduction—as a strategy to address these issues. Two applications are explored: Augmented Cognition (AugCog) and single-trial Event-Related Potential (ERP) detection. For AugCog, novel linear and nonlinear feature selection methods using Mutual Information and a statistical similarity-based extraction approach are developed. For ERP detection, comparative analyses of feature techniques across time, frequency, and spatial domains are presented. Experimental results demonstrate improved BCI performance over baseline systems, and guidelines for selecting appropriate methods based on data structure are provided.