Marios Petrov

Date of Award

2024

Document Type

Thesis

Degree Name

Bachelors

Department

Natural Sciences

First Advisor

Hamid, Fahmida

Area of Concentration

Computer Science and Neuroscience

Abstract

This research utilizes an evolutionary approach to optimizing Motor Imagery Brain-Computer Interface (MI-BCI) classification by integrating Binary Particle Swarm Optimization (BPSO) withWavelet Packet Decomposition (WPD) and Common Spatial Pattern (CSP) feature extraction methods (WCSP+BPSO). The study aims to reduce the reliance on extensive user-specific training data, a significant barrier to the practical deployment of BCIs. By applying BPSO after the WCSP feature extraction process, we develop a method that enhances classifier performance, even with limited training samples. Furthermore, through the analysis of the extracted BPSO filters, new insights into the nuances of MI-EEG data amongst people. The efficacy of this integrated approach is evaluated across three BCI Competition datasets, demonstrating that it matches or surpasses state-of-the-art results in MI-EEG classification with significantly fewer training samples. This advancement not only contributes to the field of EEG classification by providing a more efficient and practical approach to developing MI-BCIs but also has important implications for the broader accessibility and applicability of BCI technologies, particularly for individuals with mobility impairments or neurodegenerative diseases. The study underlines the potential of combining evolutionary optimization techniques with advanced signal processing methods to improve the performance of BCIs. Our method achieved averaged 5-CV accuracies of 99.5%, 97.5%, and 97.9% on the BCI Competition datasets III-IVa, IVI, and IV-IIa respectively. While our approach signals a significant methodological leap, it acknowledges the computational challenges and the need for validation in more diverse real-world settings as primary limitations.

Download

Share

COinS