Deepfake Detection Using Machine Learning Algorithms

Abstract

Deepfake, a new video manipulation technique, has drawn much attention recently. Among the unlawful or nefarious applications, Deepfake has been used for spreading misinformation, fomenting political discord, smearing opponents, or even blackmailing. As the technology becomes more sophisticated and the apps for creating them ever more available, detecting Deepfake has become a challenging task, and accordingly researchers have proposed various deep learning (DL) methods for detection. Though the DL-based approach can achieve good solutions, this paper presents the results of our study indicating that traditional machine learning (ML) techniques alone can obtain superior performance in detecting Deepfake. The ML-based approach is based on the standard methods of feature development and feature selection, followed by training, tuning, and testing an ML classifier. The advantage of the ML approach is that it allows better understandability and interpretability of the model with reduced computational cost. We present results on several Deepfake datasets that are obtained relatively fast with comparable or superior performance to the state-of-the-art DL-based methods: 99.84% accuracy on FaceForecics++, 99.38% accuracy on DFDC, 99.66% accuracy on VDFD, and 99.43% on Celeb-DF datasets. Our results suggest that an effective system for detecting Deepfakes can be built using traditional ML methods.