Hybrid AI Approach Combining Adversarial Deep Learning Models For Deepfake Detection And Enhanced Digital Forensics Verification

Abstract

With the swift development of AI-generated content, deepfake media has become an impending danger to digital security and trust. The hyper-realistic manipulated videos present severe challenges in applications like journalism, law enforcement, and social media where authenticity matters. Thus, the aim of this work is to construct a reliable detection system that would give a high percentage of true detection and can be applied to various kinds of deepfakes. For this purpose, this paper employs a fused CNNs (Convolutional Neural Networks) and GANs (Generative Adversarial Networks) where CNNs are effective in feature learning aspect while GANs specializes in detecting anomalies. The training and testing of the model was conducted on FaceForensics++ using TensorFlow for preprocessing and designing the model The proposed model built from the previous models to increase the accuracy of the model up to 99.80, precision of 99.90%, recall of 99.70 and F1-score of 99.80%. These observations indicate that one could achieve the integration of adversarial learning alongside deep convolutional frameworks in the detection of subtle manipulation models embedded in facial images. In other words, this coupled methodology can effectively establish a method for identifying deepfake content and pave the way for safer digital forensic authentication.