The emergence of advanced synthetic media, particularly deepfakes generated via Generative Adversarial Networks (GANs) and diffusion models, has created a critical demand for forensic detection models capable of generalizing across diverse manipulation methods. Conventional convolutional neural networks (CNNs), while achieving high accuracy on closed-set benchmarks, exhibit a significant “generalization gap” when exposed to novel forgery techniques or low-quality social media content. This paper presents the final empirical evaluation and comprehensive performance analysis of the Meta-learning EfficientNet Vision Transformer (MEViT) framework—a hybrid architecture that integrates EfficientNet for local texture feature extraction with Vision Transformers (ViT) for global context modelling. The optimization strategy employs Pair-Discrimination Loss (PDL) and Domain Adjustment Loss (DAL) within an episodic meta-learning schedule to bridge the generalization gap. Extensive experiments on FaceForensics++ (FF++) and Celeb-DF benchmarks demonstrate that MEViT achieves 98.4% average detection accuracy on FF++ (c23) and maintains a strong AUC of 89.2% on the unseen Celeb-DF dataset—surpassing Xception, EfficientNet-B7, and Multi-Domain Transformer baselines by significant margins. Ablation studies confirm the indispensable contribution of each architectural component, and comparative analyses with multimodal systems validate the competitiveness of the visual-only MEViT approach. Explainability analysis via Grad-CAM further demonstrates that MEViT correctly localizes forensic artifacts in facial regions. These results establish MEViT as a robust, generalizable, and practically deployable solution for next-generation digital forensics.

{{conclusion}}

[1]R. Wagh, S. Thorat, R. Pohakar, S. Y. Mandlik, and A. A. Khatri, “AI-powered detection of deepfakes using EfficientNet and Vision Transformer,” Int. J. Adv. Res. Sci. Commun. Technol. (IJARSCT), vol. 6, no. 9, Nov. 2025. [2]V.-N. Tran, H.-S. Le, P. Choi, S.-H. Lee, and K.-R. Kwon, “MEViT: Generalization of deepfake detection with meta-learning EfficientNet Vision Transformer,” IEEE Open J. Comput. Soc., vol. 6, pp. 104–118, May 2025. [3]Q. Man, S.-J. Gee, and Y.-I. Cho, “Multi-domain perception transformer for generalized forgery image detection,” Appl. Sci., vol. 16, no. 1, Art. no. 533, Dec. 2025. [4]K. L. Shu and M.-J.-S. Wang, “Multi-domain feature fusion transformer with cross-domain robustness for facial expression recognition,” Symmetry, vol. 17, no. 1, Art. no. 88, Dec. 2025. [5]T. Tong and D. Anastasiu, “Deepfake detection using spatiotemporal methods and vision-language models,” in Proc. 31st ACM SIGKDD (KDD ’25), Aug. 2025, pp. 1–12. [6]A. Yermakov, J. Cech, J. Matas, and M. Fritz, “Deepfake detection that generalizes across benchmarks,” arXiv:2508.06248, Aug. 2025. [7]F. Shaad, “Multi-modal deepfake detection: Analyzing video, audio, and text for enhanced forgery identification,” ResearchGate, Jan. 2025. [8]L. K. Joshi and S. Joshi, “Deepfake detection using multimodal AI,” Int. J. Res. Innov. Appl. Sci. (IJRIAS), vol. 10, no. 5, pp. 355–357, May 2025. [9]M. Wang, “Deepfake detection: A multimodal survey,” ITM Web Conf., vol. 78, Art. no. 02027, 2025. [10]S. AlMuhaideb, H. Alshaya, L. Almutairi, D. Alomran, and S. T. Alhamed, “LightFakeDetect: A lightweight model for deepfake detection in videos that focuses on facial regions,” Mathematics, vol. 13, no. 19, Art. no. 3088, Sep. 2025. [11]S. Peng et al., “Wmamba: Wavelet-based Mamba for face forgery detection,” in Proc. 33rd ACM Int. Conf. Multimedia (MM ’25), Oct. 2025. [12]N. Mansoor and A. I. Iliev, “Explainable AI for deepfake detection,” Appl. Sci., vol. 15, no. 2, Art. no. 725, Jan. 2025. [13]H. Qian et al., “From black boxes to glass boxes: Explainable AI for trustworthy deepfake forensics,” Cryptography, vol. 9, no. 4, Art. no. 61, Dec. 2025. [14]X. Hu, “A comprehensive evaluation of deepfake detection methods: Approaches, challenges and future prospects,” ITM Web Conf., vol. 73, Art. no. 03002, 2025. A. Ashraf Bekheet, A. S. Ghoneim, and G. Khoriba, “Unmasking the digital deception: A comprehensive survey of large vision models for deepfake detection,” Inform. Bull., vol. 7, no. 2, 2025.

© 2025 International Journal of Engineering and Techniques (IJET).

Digital Object Identifier (DOI)

IJET assigns a unique DOI to every accepted article via Zenodo for persistent linking and citation. Your article’s DOI: https://doi.org/{{doi}}

Open DOI About Zenodo DOI

Close