A Novel Deep Learning Framework for DeepFake Detection Using Attention-Enhanced EfficientNet-B7

Koyluoglu, Emin Cem, Staikopoulos, Athanasios and Raj, Kislay (2026) A Novel Deep Learning Framework for DeepFake Detection Using Attention-Enhanced EfficientNet-B7. In: Artificial Intelligence and Cognitive Science. AICS 2025. Communications in Computer and Information Science (2950). Springer, Cham, Dublin, Ireland, pp. 401-413. ISBN 978-303225808-3

Full text not available from this repository.

Abstract

DeepFakes are AI-generated synthetic media that manipulate facial identities to create hyper-realistic but fake content, posing serious threats to privacy, trust, and digital security. Traditional DeepFake detection models, such as basic CNNs or transfer learning using VGG16 and Xception, often lack the precision to identify subtle facial manipulations due to limited attention mechanisms and insufficient feature localization. To address these limitations, this study proposes a novel deep learning framework combining EfficientNet-B7 with a Custom Soft Spatial Attention Mechanism. This attention block enables the model to focus on critical facial regions most likely to exhibit manipulation artifacts, improving detection sensitivity. A comprehensive experimental setup was conducted on a balanced dataset of 20,000 images (10,000 real/10,000 fake), comparing models trained using Adam and SGD optimizers, with and without augmentation. The proposed EfficientNet-B7 with attention achieved 97% accuracy, outperforming baseline CNN (50%), VGG16 (78%), Xception (65%), MobileNetV2 (96%), and standard EfficientNet-B7 (96%). The framework includes early stopping, learning rate scheduling, and LIME-based explainability. This research delivers a scalable, accurate, and interpretable solution for DeepFake detection, advancing AI-driven media forensics.

Item Type:	Book Section
Uncontrolled Keywords:	DeepFake Detection; EfficientNet-B7; Explainable AI; Image Forensics; Spatial Attention; Transfer Learning
Subjects:	Q Science > QH Natural history > QH301 Biology > Methods of research. Technique. Experimental biology > Data processing. Bioinformatics > Artificial intelligence > Computer vision Q Science > Q Science (General) > Self-organizing systems. Conscious automata > Artificial intelligence > Computer vision B Philosophy. Psychology. Religion > BJ Ethics > Conduct of life > Reliability > Information integrity > Data integrity Q Science > Q Science (General) > Self-organizing systems. Conscious automata > Machine learning
Divisions:	School of Computing > Staff Research and Publications
Depositing User:	Tamara Malone
Date Deposited:	18 Sep 2026 15:32
Last Modified:	18 Sep 2026 15:32
URI:	https://norma.ncirl.ie/id/eprint/9935

Actions (login required)

View Item