DeepFake Detection Using Deep Learning

Originating in 2017, DeepFakes utilizes artificial intelligence algorithms, particularly Generative Adversarial Networks (GANs), to generate deceptive videos and images. Despite the wide-ranging applications of GANs, their potential misuse raises ethical concerns. The rapid advancement of technology in multimedia has led to the emergence of malicious activities, such as DeepFake manipulation in politics and entertainment. As manipulated visuals become increasingly realistic, the need for a DeepFake detection approach becomes essential. The research employs two distinct Convolutional Neural Network architectures for classification, emphasizing the escalating importance of DeepFake detection in contemporary society. In this research, DeepFake detection techniques are developed by building two CNN models based on ResNet50 and DenseNet121 architectures. A comparative analysis is done for the two models based on accuracy and F1-score efficiency parameters. Experimental results indicate that the CNN models employed in the study achieve an accuracy of 86% and an F1-score of 0.87, respectively.