Multimodal Deepfake Detection Using Transformer-Based Large Language Models: A Path Toward Secure Media and Clinical Integrity
Kutub Thakur, Md Abu Sayed, Sanjida Akter Tisha, M. K. Alam, Md Tarek Hasan, Jannatul Ferdous Shorna, Sadia Afrin, Md Zahin Hossain George, Eftekhar Hossain Ayon · The American Journal of Engineering And Technology · 2025
Deepfakes pose a significant threat across various domains by generating highly realistic manipulated audio-visual content, with critical implications for security and clinical environments. This paper presents a robust multimodal deepfake detection framework powered by transformer-based large language models (LLMs) that effectively analyze and integrate visual, auditory, and textual modalities. Utilizing the FakeAVCeleb dataset, we compare our proposed model with traditional machine learning and deep learning methods, including Logistic Regression, Support Vector Machine (SVM), Random Forest, and Long Short-Term Memory (LSTM) networks. Experimental results demonstrate that the transformer-based model significantly outperforms others, achieving an accuracy of 96.55%, precision of 96.47%, recall of 96.50%, F1-score of 96.48%, and an AUC of 0.97. This enhanced performance is attributed to the model’s ability to capture complex semantic and temporal dependencies across modalities. The findings suggest the proposed model’s strong potential for real-world applications such as telemedicine, clinical video authentication, and digital identity verification, establishing a promising direction for deploying deepfake detection technologies in sensitive and high-stakes environments.