Comprehensive Evaluation of Deepfake Detection Models: Accuracy, Generalization, and Resilience to Adversarial Attacks

The rise of deepfakes—synthetic media generated using artificial intelligence—threatens digital content authenticity, facilitating misinformation and manipulation. However, deepfakes can also depict real or entirely fictitious individuals, leveraging state-of-the-art techniques such as generative ad...

Full description

Saved in:
Bibliographic Details
Main Authors: Maryam Abbasi, Paulo Váz, José Silva, Pedro Martins
Format: Article
Language:English
Published: MDPI AG 2025-01-01
Series:Applied Sciences
Subjects:
Online Access:https://www.mdpi.com/2076-3417/15/3/1225
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1850199428709744640
author Maryam Abbasi
Paulo Váz
José Silva
Pedro Martins
author_facet Maryam Abbasi
Paulo Váz
José Silva
Pedro Martins
author_sort Maryam Abbasi
collection DOAJ
description The rise of deepfakes—synthetic media generated using artificial intelligence—threatens digital content authenticity, facilitating misinformation and manipulation. However, deepfakes can also depict real or entirely fictitious individuals, leveraging state-of-the-art techniques such as generative adversarial networks (GANs) and emerging diffusion-based models. Existing detection methods face challenges with generalization across datasets and vulnerability to adversarial attacks. This study focuses on subsets of frames extracted from the DeepFake Detection Challenge (DFDC) and FaceForensics++ videos to evaluate three convolutional neural network architectures—XCeption, ResNet, and VGG16—for deepfake detection. Performance metrics include accuracy, precision, F1-score, AUC-ROC, and Matthews Correlation Coefficient (MCC), combined with an assessment of resilience to adversarial perturbations via the Fast Gradient Sign Method (FGSM). Among the tested models, XCeption achieves the highest accuracy (89.2% on DFDC), strong generalization, and real-time suitability, while VGG16 excels in precision and ResNet provides faster inference. However, all models exhibit reduced performance under adversarial conditions, underscoring the need for enhanced resilience. These findings indicate that robust detection systems must consider advanced generative approaches, adversarial defenses, and cross-dataset adaptation to effectively counter evolving deepfake threats.
format Article
id doaj-art-b7ff80e3195242b8a3a9d792739dffef
institution OA Journals
issn 2076-3417
language English
publishDate 2025-01-01
publisher MDPI AG
record_format Article
series Applied Sciences
spelling doaj-art-b7ff80e3195242b8a3a9d792739dffef2025-08-20T02:12:37ZengMDPI AGApplied Sciences2076-34172025-01-01153122510.3390/app15031225Comprehensive Evaluation of Deepfake Detection Models: Accuracy, Generalization, and Resilience to Adversarial AttacksMaryam Abbasi0Paulo Váz1José Silva2Pedro Martins3Applied Research Institute, Polytechnic of Coimbra, 3045-093 Coimbra, PortugalResearch Center in Digital Services (CISeD), Polytechnic of Viseu, 3504-510 Viseu, PortugalResearch Center in Digital Services (CISeD), Polytechnic of Viseu, 3504-510 Viseu, PortugalResearch Center in Digital Services (CISeD), Polytechnic of Viseu, 3504-510 Viseu, PortugalThe rise of deepfakes—synthetic media generated using artificial intelligence—threatens digital content authenticity, facilitating misinformation and manipulation. However, deepfakes can also depict real or entirely fictitious individuals, leveraging state-of-the-art techniques such as generative adversarial networks (GANs) and emerging diffusion-based models. Existing detection methods face challenges with generalization across datasets and vulnerability to adversarial attacks. This study focuses on subsets of frames extracted from the DeepFake Detection Challenge (DFDC) and FaceForensics++ videos to evaluate three convolutional neural network architectures—XCeption, ResNet, and VGG16—for deepfake detection. Performance metrics include accuracy, precision, F1-score, AUC-ROC, and Matthews Correlation Coefficient (MCC), combined with an assessment of resilience to adversarial perturbations via the Fast Gradient Sign Method (FGSM). Among the tested models, XCeption achieves the highest accuracy (89.2% on DFDC), strong generalization, and real-time suitability, while VGG16 excels in precision and ResNet provides faster inference. However, all models exhibit reduced performance under adversarial conditions, underscoring the need for enhanced resilience. These findings indicate that robust detection systems must consider advanced generative approaches, adversarial defenses, and cross-dataset adaptation to effectively counter evolving deepfake threats.https://www.mdpi.com/2076-3417/15/3/1225deepfakesdeep learningXCeptionResNetVGGDFDC
spellingShingle Maryam Abbasi
Paulo Váz
José Silva
Pedro Martins
Comprehensive Evaluation of Deepfake Detection Models: Accuracy, Generalization, and Resilience to Adversarial Attacks
Applied Sciences
deepfakes
deep learning
XCeption
ResNet
VGG
DFDC
title Comprehensive Evaluation of Deepfake Detection Models: Accuracy, Generalization, and Resilience to Adversarial Attacks
title_full Comprehensive Evaluation of Deepfake Detection Models: Accuracy, Generalization, and Resilience to Adversarial Attacks
title_fullStr Comprehensive Evaluation of Deepfake Detection Models: Accuracy, Generalization, and Resilience to Adversarial Attacks
title_full_unstemmed Comprehensive Evaluation of Deepfake Detection Models: Accuracy, Generalization, and Resilience to Adversarial Attacks
title_short Comprehensive Evaluation of Deepfake Detection Models: Accuracy, Generalization, and Resilience to Adversarial Attacks
title_sort comprehensive evaluation of deepfake detection models accuracy generalization and resilience to adversarial attacks
topic deepfakes
deep learning
XCeption
ResNet
VGG
DFDC
url https://www.mdpi.com/2076-3417/15/3/1225
work_keys_str_mv AT maryamabbasi comprehensiveevaluationofdeepfakedetectionmodelsaccuracygeneralizationandresiliencetoadversarialattacks
AT paulovaz comprehensiveevaluationofdeepfakedetectionmodelsaccuracygeneralizationandresiliencetoadversarialattacks
AT josesilva comprehensiveevaluationofdeepfakedetectionmodelsaccuracygeneralizationandresiliencetoadversarialattacks
AT pedromartins comprehensiveevaluationofdeepfakedetectionmodelsaccuracygeneralizationandresiliencetoadversarialattacks