Evaluating Contrastive Learning Approaches for Robust P300 Detection in Noisy EEG Environments

Quarni, Muhammad Awais (2025) Evaluating Contrastive Learning Approaches for Robust P300 Detection in Noisy EEG Environments. Masters thesis, Dublin, National College of Ireland.

Preview	PDF (Master of Science) Download (2MB) | Preview
Preview	PDF (Configuration Manual) Download (752kB) | Preview

Abstract

The extraction of the P300 event-related potential (ERP) electroencephalogram (EEG) data is critical for the development of robust brain-computer interface (BCI) systems, especially in realistic, noisy environments. The study investigates the effectiveness of modern contrastive self-supervised learning methods (SimCLR, TS-TCC, and VICReg) in comparison to a conventional supervised convolutional neural network (CNN) for single-trial P300. The experiments were conducted on the AMBER dataset, a multi-subject, multi-session EEG corpus explicitly collected under both clean and ecologically valid noisy conditions. Due to computational constraints, the analysis was limited to the RSVP tasks X1(clean) and X4 (noisy) from all subjects. EEG epochs were preprocessed, normalized, and augmented, and feature representations were learned via contrastive pretraining. The resulting embeddings were evaluated using downstream classifiers with model performance assessed by accuracy, ROC-AUC, and F1-Score for the rare target (P300) class. Results show that all models struggled to robustly detect P300 targets under real-world noise and data imbalance, with low recall and ROC-AUC across the board. Among the contrastive models, VICReg achieved the highest recall (0.556) and F1-score (0.225) for targets. This finding suggests that variance-invariance-covariance regularisation may enhance representation learning for challenging EEG tasks, but also underscores the persistent difficulty of P300 detection in noisy, resource-limited scenarios. This study provides a transparent benchmark and highlights key challenges for future research with more comprehensive analysis, including all tasks and a larger sample size as well as advanced augmentation and computational strategies are recommended to further advance robust P300 detection in real-world BCI applications.

Item Type:	Thesis (Masters)
Supervisors:	Name Email Rifai, Hicham UNSPECIFIED
Subjects:	Q Science > QA Mathematics > Electronic computers. Computer science T Technology > T Technology (General) > Information Technology > Electronic computers. Computer science R Medicine > RC Internal medicine > RC0321 Neuroscience. Biological psychiatry. Neuropsychiatry
Divisions:	School of Computing > Master of Science in Data Analytics
Depositing User:	Ciara O'Brien
Date Deposited:	26 Aug 2026 10:28
Last Modified:	26 Aug 2026 10:28
URI:	https://norma.ncirl.ie/id/eprint/9657

Actions (login required)

View Item