Copyright © 2026 Authors retain the copyright of this article. This article is an open access article distributed under the Creative Commons Attribution License which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.

@article{209546,
        author = {Nandini Kalshetty and Samiksha Kharosekar and Shreya Rokade and Mrs. Punam Patil},
        title = {A Systematic Review of Deepfake Detection Techniques: Cross-Dataset Generalization, Computational Efficiency, And Performance Evaluation},
        journal = {International Journal of Innovative Research in Technology},
        year = {2026},
        volume = {13},
        number = {no},
        pages = {795-798},
        issn = {2349-6002},
        url = {https://ijirt.org/article?manuscript=209546},
        abstract = {Deepfake technology has emerged as a major challenge to the authenticity and trustworthiness of digital media. The rapid development of artificial intelligence and deep learning has made it possible to generate highly realistic manipulated images, videos, and other forms of digital content. As a result, reliable deepfake detection systems are increasingly important for digital forensics, cybersecurity, social media platforms, and the protection of information integrity. This paper presents a systematic review of deepfake detection techniques published between 2018 and 2025, with particular attention to deep learning, traditional machine learning, and classical image-processing approaches. 
The review focuses on detection performance, computational efficiency, and cross-dataset generalization. Studies evaluated using FaceForensics++, the DeepFake Detection Challenge (DFDC), and Celeb-DF indicate that Transformer-based methods generally demonstrate stronger cross-dataset generalization than evaluated CNN-based approaches. However, this improved generalization comes with substantially higher computational requirements.
Traditional machine learning approaches, particularly Random Forest, can provide competitive performance in some settings while requiring fewer computational resources. Overall, performance degradation of approximately 10–15% across datasets remain a significant challenge. These findings indicate that future deepfake detection research should prioritize robustness, generalization, computational efficiency, and real-world deployment rather than focusing exclusively on accuracy.},
        keywords = {Deepfake; Deepfake Detection; Machine Learning; Deep Learning; Computer Vision; Cross Dataset Generalization; Digital Forensics; Artificial Intelligence},
        month = {October},
        }