An Efficient Deepfake Video Detection Approach with Combination of EfficientNet and Xception Models Using Deep Learning
Serhat Ataş, Ismail Ilhan, Mehmet Karakse
Abstract
Artificial intelligence is used in many areas and is constantly being developed. In recent years, videos made with deep fakes, which are often heard, have also developed. The use of videos made with deep fakes as blackmail in people's lives, manipulating the videos of important people to cause anxiety on people and etc. due to the fact that it poses a threat in many areas presents a big problem today. Efforts are being made to prevent this threat by detecting deep fake videos. Deep fake detection is still not fully resolved. For this reason, prominent technology companies provide support to researchers in this field and develop deep fraud detection by suggesting methods and organizing contests on most platforms such as Kaggle. In this article, a detection method is proposed to minimize the current concern of deep forgery. In the proposed method, the Xception model with high performance and speed and the EfficientNetB4 model with high accuracy were used. The proposed method aims to achieve better results and improvements in detecting fake videos.