Date of Award

2-4-2025

Publication Type

Thesis

Degree Name

M.A.Sc.

Department

Electrical and Computer Engineering

Keywords

Additive Manufacturing; Genetic Algorithm; Heuristics; Logistics; Machine Learning; C26Supply Chain Management

Supervisor

Esam Abdel-Raheem

Supervisor

Luis Rueda

Rights

info:eu-repo/semantics/openAccess

Abstract

This thesis presents a comprehensive study to enhance the accuracy and efficiency of steady-state visual evoked potential (SSVEP) detection in brain-computer interface (BCI) systems. The research leverages advanced feature extraction techniques, including Power spectral density analysis (PSDA), canonical correlation analysis (CCA), filter bank canonical correlation analysis (FBCCA), and fusing canonical correlation analysis (FoCCA), combined with feature selection methods such as mutual information (MI), fisher score (FS), and recursive feature elimination (RFE). These methodologies are systematically evaluated to determine their effectiveness in isolating relevant features for accurate classification of SSVEP signals, with a focus on optimizing the feature set to improve classification performance and minimize computational complexity. The dataset utilized in this research is the publicly available Kalunga et al. (2013) SSVEP dataset, consisting of electroencephalogram (EEG) recordings from four subjects in response to visual stimuli at frequencies of 13 Hz, 17 Hz, and 21 Hz. Preprocessing steps include the application of a notch filter to remove line noise, common average referencing (CAR) to reduce channel noise, bandpass filtering to isolate relevant frequencies, and fast independent component analysis (Fast ICA) algorithm for further noise reduction. The preprocessed data is subjected to the feature extraction and selection techniques and classified using machine learning algorithms such as support vector machine (SVM) with RBF kernel (SVM-RBF), linear discriminant analysis (LDA), multi-layer perceptron (MLP), k-nearest neighbors (kNN), and convolutional neural network (CNN). The results highlight FBCCA as the most effective feature extraction method, particularly when combined with MI or RFE for feature selection. Among the classifiers, CNN emerges as the most robust and reliable, demonstrating superior performance across various metrics when paired with FBCCA, Fast ICA, and feature selection techniques. The combination of FBCCA, Fast ICA, RFE or MI, and CNN is established as the optimal pipeline for SSVEP-based BCI applications, providing a scalable and reliable solution for real-time systems. These findings contribute to advancing BCI technology, enabling the development of assistive and communication devices, and setting a benchmark for future research in the field.

Download

Share

COinS