Answers are grounded in this paper and cite their sources.
Quick developments in generative artificial intelligence (GenAI) have ushered in an era where highly realistic manipulated videos are now being produced at an unprecedented pace, which has thrown a big challenge to the authenticity of digital media, cybersecurity, public trust, and information integrity online. Most of the prior works on deepfake detection have been dedicated to either spatial feature extraction or temporal sequence modeling, and fall short in addressing the increasingly powerful deep learning-based deepfake samples. To overcome such drawbacks, this paper introduces an Attention-Enhanced Hybrid Deep Learning Framework for Robust Deepfake Video Detection and Analysis, which combines EfficientNetB2 to extract discriminative spatial features and a Bidirectional Gated Recurrent Unit (Bi-GRU) with the inclusion of an Attention Mechanism to learn temporal features adaptively. The proposed framework is able to effectively account for visual artifacts at a frame level and long-range temporal inconsistencies within manipulated videos. In addition, three universal architectures, namely Hybrid Transformer, Temporal Pooling, and CNN3D, are used as baseline models for comparison. Here, experiments are performed on a balanced set of 2,952 authentic and manipulated videos, using a uniform frame sampling, a uniform image normalization, and a uniform preprocessing. Class-weighted loss with the AdamW optimizer is used to train the models and evaluate the models with accuracy, precision, recall, F1-score, and Area Under the Receiver Operating Characteristic Curve (AUC). The results of the experiments presented in this paper demonstrate that the proposed Hybrid_GRU_Attention model can attain the highest detection performance with high accuracy of 95.75% and high AUC of 0.9771, which are better than the existing baseline models with high computational efficiency. The suggested solution gives a trustworthy and scalable choice in multimedia forensics, digital content validation, and future deepfake detection applications.
No takes yet. Share an insight, caveat, or question.
Prasad et al. (2026) studied this question.
Synapse has enriched 10 related papers. Consider them for comparative context: