Juhi Singh,
Shreya Saxena,
A. Madhav Ainesh,
Yanshi Goyal,
- Student, Department of Computer Science and Engineering, Amity School of Engineering and Technology, Amity University Noida, Uttar Pradesh, India
- Associate Professor, Department of Computer Science and Engineering, Amity School of Engineering and Technology, Amity University Noida, Uttar Pradesh, India
- Student, Department of Computer Science and Engineering, Amity School of Engineering and Technology, Amity University Noida, Uttar Pradesh, India
- Student, Department of Computer Science and Engineering, Amity School of Engineering and Technology, Amity University Noida, Uttar Pradesh, India
Abstract
Deepfake technology presents non-exemplary threats to video conferencing platforms, enabling advanced fraud, impression and misinformation campaigns worth billions annually. Current detection methods either exhibit latencies exceeding 100ms or rely on server-side cloud processing, raising privacy concerns. This paper presents DeepConfGuard, a lightweight hybrid architecture combining MobileNetV2 for spatial feature extraction, a bidirectional LSTM with attention for temporal modelling, and EfficientNetV2 for refinement. It reaches 94.8% accuracy with 85 ms end‑to‑end latency on about 95,000 video samples, which is suitable for real‑time deployment on edge devices. Key contributions are: (1) multi-scale CNN LSTM fusion detecting both individual frame artifacts and motion anomalies, (2) SHAP-based pixel-level explainability via heatmap overlays building operator trust, (3) Flask-SocketIO inference server enabling 30fps local processing with GDPR compliance, (4) comprehensive cross-dataset validation showing 91.2% generalization on unseen FaceForensics++and adversarial strongness improvements via knowledge distillation. Ablation studies quantify LSTM’s +1.3% accuracy contribution and attention’s +0.9% uplift. While maintaining detection performance and safeguarding critical meeting content, the suggested system lessens reliance on external servers. It is useful for privacy-conscious video communication environments because of its small size, which enables effective inference under constrained computer resources. The findings show a fair trade-off between deployment efficiency, interpretability, processing speed, and detection accuracy.
Keywords: Deepfake detection, Video conferencing security, MobileNetV2, LSTM, SHAP explainability, Real-time inference, FaceForensics++, Celeb-DF, DFDC, adversarial strongness
References
- Lin LS. Examining the role of deepfake technology in organized fraud: Legal, security, and governance challenges. Frontiers in Law. 2025 Apr 16;4:6-17.
- Pedersen KT, Pepke L, Stærmose T, Papaioannou M, Choudhary G, Dragoni N. Deepfake-driven social engineering: Threats, detection techniques, and defensive strategies in corporate environments. Journal of Cybersecurity and Privacy. 2025 Apr 27;5(2):18.
- Khan AA, Laghari AA, Inam SA, Ullah S, Shahzad M, Syed D. A survey on multimedia-enabled deepfake detection: state-of-the-art tools and techniques, emerging trends, current challenges & limitations, and future directions. Discover Computing. 2025 Apr 25;28(1):48.
- Chesney R, Citron D. Deepfakes and the new disinformation war. Foreign affairs. 2019 Jan 1;98(1):147-55.
- Fletcher J. Deepfakes, artificial intelligence, and some kind of dystopia: The new faces of online post-fact performance. Theatre journal. 2018;70(4):455-71.
- Sharma R, Sharma A. Deepfake Video Detection: A Comprehensive Deep Learning-Based Approach. In2025 14th International Conference on System Modeling & Advancement in Research Trends (SMART) 2025 Nov 14 (pp. 30-36). IEEE.
- Vaswani A, Shazeer N, Parmar N, Uszkoreit J, Jones L, Gomez AN. Attention is all you need [J]. Advances in neural information processing systems. 2017 Dec 4;30(1):261-72.
- He K, Zhang X, Ren S, Sun J. Deep residual learning for image recognition. InProceedings of the IEEE conference on computer vision and pattern recognition 2016 (pp. 770-778).
- Gerstner CR, Farid H. Detecting real-time deep-fake videos using active illumination. In2022 IEEE/CVF Conference on Computer Vision and Pattern Recognition Workshops (CVPRW) 2022 Jun 19 (pp. 53-60). IEEE.
International Journal of Electronics Automation
| Volume | 04 | |
| 02 | ||
| Received | 24/07/2026 | |
| Accepted | 05/08/2026 | |
| Published | 12/08/2026 | |
| Publication Time | 19 Days |