Deepfake Detection using Hybrid Deep Learning: Enhancing Accuracy with ResNet, GANs, and Attention Mechanisms

Deep fakes are becoming increasingly realistic because to the quick development of artificial intelligence, especially with the use of Generative Adversarial Networks (GANs). This presents serious threats to security and privacy. Since deepfakes can be used maliciously to distort photos, movies, and sounds, it is imperative that they are detected as soon as possible. In this paper, a hybrid deep learning model that combines the advantages of Residual Neural Networks (RESNET) and Generative Adversarial Networks (GANs) with Channel-Wise Attention Mechanisms is presented as a comprehensive method for deepfake detection. The suggested model seeks to precisely detect regions of the face that have been altered and classify faces as authentic or modified. To focus on important facial traits, the model combines the generative capacity of GANs with the discriminative skills of RESNET, which are enhanced by Channel Wise Attention Mechanisms. DFDC, FaceForensic++, CelebDF, and other benchmark datasets are used to assess the performance of this hybrid model. With a precision of 0.79, recall of 0.88, F1-score of 0.83, accuracy of 0.83, and ROC AUC Score of 0.825, it has high performance characteristics. Furthermore, the model undergoes testing against hostile attacks and proves its resilience in actual situations. In addition to accurate deepfake identification, our method offers comprehensive localization of changed regions, which provides important information about which specific facial features have been altered. This feature improves the model’s usefulness in cybersecurity, social media content moderation, and identity verification. This research extends the field of deepfake identification and helps to build more dependable and secure digital systems by tackling the detection as well as the localization of fake characteristics.