Detection of deepfake media files using deep learning methods

Deepfake media is any kind of audiovisual data where people's images and/or voices are changed or imitated. Deepfake media, where people's faces are changed, is more frequently encountered. Developments in artificial intelligence algorithms in parallel with technological developments have made it possible to produce much more realistic deepfake media. There are examples of deepfake media being used in many different areas, both well-intentioned and malicious. In order for individuals not to be exposed to the misuse of deepfake media technology, they should not share their visual and audio media publicly on platforms such as social media. It is also very important for states to take the necessary measures to prevent the misuse of deepfake media. In recent years, there has been an increase in the number of academic studies on the detection of deepfake media. The main subject of our study is to detect deepfake media using deep learning architectures. In this context, 5 different pre-trained models (VGG16, EfficientNetB4, DenseNet201, InceptionV3, ResNet50V2) were tested on the FaceForensics++ dataset in the Google Colab environment. EfficientNetB4 was the most successful model with an AUC value of 0,93.