Localized Deepfake Detection: Enhancing Authenticity Verification in Irish Media

Correia, Simi Silvester (2025) Localized Deepfake Detection: Enhancing Authenticity Verification in Irish Media. Masters thesis, Dublin, National College of Ireland.

Preview	PDF (Master of Science) Download (4MB) | Preview
Preview	PDF (Configuration Manual) Download (988kB) | Preview

Abstract

Deepfakes are becoming harder to detect and more convincing than ever, threatening the trust we have in news, public figures, and digital communication. While many detection models already exist, most are trained on large, global datasets that do not reflect regional differences in faces, speech, or media context. This makes them less effective when applied to local content. In this research, the researcher focuses on building a region-specific deepfake detection system for Irish media. The project was carried out in three phases. In the first phase, real and fake images from Irish news sources were collected and used to build a proof-of-concept model. In the second phase, fake images were generated using the Stable Diffusion model, while realistic non-manipulated images were taken from This Person Does Not Exist because of their close resemblance to Irish features. In the third phase, the researcher tested cross-dataset performance by training on the global DFDC dataset and testing on Irish data, where accuracy dropped to 50%, showing that global models fail to generalize well to local content. Finally, by mixing both datasets and emphasizing the Irish samples, the region-specific approach produced much stronger results. Overall, the global dataset achieved 50% accuracy (F1: 33%, ROC-AUC: 46.9%), while the region specific dataset reached 74% accuracy (F1: 76%, ROC-AUC:%). These findings suggest that localized training not only improves accuracy but also strengthens the reliability of deepfake detection in Ireland, offering a model that can be adapted by other regions to safeguard media integrity.

Item Type:	Thesis (Masters)
Supervisors:	Name Email Menghwar, Teerath Kumar UNSPECIFIED
Uncontrolled Keywords:	Deepfake Detection; Region Specific; Stable Diffusion; EfficientNet-B3; Irish Media
Subjects:	H Social Sciences > HE Transportation and Communications > Broadcasting Media Q Science > Q Science (General) > Self-organizing systems. Conscious automata > Machine learning Z Bibliography. Library Science. Information Resources > ZA Information resources > ZA4150 Computer Network Resources > The Internet > World Wide Web > Websites > Online social networks T Technology > TK Electrical engineering. Electronics. Nuclear engineering > Telecommunications > The Internet > World Wide Web > Websites > Online social networks
Divisions:	School of Computing > Master of Science in Data Analytics
Depositing User:	Ciara O'Brien
Date Deposited:	01 Jul 2026 08:43
Last Modified:	01 Jul 2026 08:43
URI:	https://norma.ncirl.ie/id/eprint/9420

Actions (login required)

View Item