Deepfake Video Detection Using Spatial and Temporal Feature Learning

This research addresses the critical challenge of synthetic media identification by proposing a robust hybrid spatial-temporal deep learning framework. By combining the feature extraction prowess of EfficientNet-B0 with the sequential modeling capabilities of Bidirectional LSTM networks, the system successfully detects deepfake videos by analyzing both spatial artifacts and temporal motion inconsistencies [10], [15]. The two-phase training methodology, combined with rigorous data augmentation and hyperparameter tuning, yields a highly accurate model capable of generalizing to unseen data, ultimately achieving an optimized F1-score of 0.98. Future advancements should focus on domain adaptation techniques to improve cross-dataset reliability [17], the inte-gration of Explainable Artificial Intelligence (XAI) to provide interpretability for the model’s decisions [7], and the develop-ment of lightweight mobile deployment models for real-time analysis. Additionally, exploring multimodal detection systems that combine visual features with audio and physiological signals [8] will provide a more comprehensive defense against increasingly sophisticated generative models.

[1]     V. Ahire, A. Kulkarni, and S. Patil, “SFANet: Spatial-Frequency Atten-tion Network for Deepfake Detection,” arXiv preprint, 2025. [2]     M. Alsboui et al., “Threats of Generative AI and Countermeasures,” Journal of Cybersecurity, 2024. [3]     Y. Chen et al., “Towards Generalizable Deepfake Detection with Spatial-Frequency Collaborative Learning,” International Journal of Computer Vision, 2025. [4]     G. H. Ishrak et al., “Analysis of Deepfake Modalities,” arXiv preprint, 2024. [5]     N. Jain et al., “Detecting Deepfakes: Exploring Machine Learning Models for Audio, Video, and Image Analysis,” International Journal of Intelligent Systems and Applications in Engineering, 2024. [6]     W. A. Jbara, N. A.-H. K. Hussein, and J. H. Soud, “Deepfake Detection in Video and Audio Clips: A Comprehensive Survey and Analysis,” Mesopotamian Journal of Cybersecurity, vol. 4, no. 3, pp. 233-250, 2024. [7]     J. Kaur et al., “Attention Mechanisms in Video Forensics,” Applied Intelligence, 2026. [8]     H. Khalid, S. Tariq, M. Kim, and S. S. Woo, “FakeAVCeleb: A Novel Audio-Video Multimodal Deepfake Dataset,” arXiv preprint arXiv:2108.05080, 2021. [9]     C. Y. Lin et al., “Deepfake Detection Frameworks,” MDPI Electronics, 2024. [10]    U. Masud et al., “Spatial-Temporal Artifact Analysis,” Signal Image Video Processing, 2023.