Paper Title
Deep Fake Detection to Prevent False Propagandas Using Ml Algorithms: A Comparative Study

Abstract
Manipulation of images and videos has been made easier by deepfake technology. This manipulated content is almost indistinguishable from ordinary vision. Many algorithms have been developed and are still being developed every day to detect deepfake content. Approaches followed by these algorithms may include visual, local, deep, and temporal feature-based approaches. The primary debacle faced by the researchers in coming up with deepfake detection algorithms is the changing dynamics of the models that are used to generate this deepfake content, which keeps making deepfake videos closer to real life. In this paper, we have very meticulously analysed the working mechanisms of the three main models for deepfake detection, namely ResNeXt50 + LSTM, EfficientNet, and MesoNet. The algorithms are compared on various performance measures, such as accuracy, precision, recall, and F1 scores. Conclusively, EfficientNet, with an accuracy of 97.5 % surpassed all the other algorithms in deepfake detection. Keywords - Deepfake detection, LSTM, ResNeXt, EfficientNet, MesoNet