Enhancing Deepfake Detection with Ensemble Learning: A Study on Celeb DF Dataset
Kokou Elvis Khorem Blitti, Fitsum Getachew Tola, Reshma Sunil, Anjali Diwan · 2024
Deepfakes, generated digital content using deep learning methods, have become more sophisticated nowadays with the development of new techniques like diffusion models. They are now causing a serious threat to the social balance, reducing trust in digital content. While research is still going on to improve the quality of generated content, techniques are also being developed to detect fake content. This research tackles particularly deepfake images. This research develops and tests a robust model based on ensemble learning on the Celeb-DF dataset. The ensemble technique used in this study is the averaging technique. At the same time, the efficiency of transfer learning models in detecting deepfakes is assessed. An accuracy of 97.62 % was achieved, testifying to the power of transfer learning and ensemble models. This paper showcases how different models can be used together to improve the accuracy and sensitivity of the models created to tackle the deepfake detection problem.