A Comprehensive Review of DeepFake Detection Methods Using Machine Learning

Presenter (s)

Sonia Sherif, Hawi Atinafu, Bethel Zegeye

Abstract or Description

Deepfakes are synthetic media created using deep learning techniques, typically involving the manipulation of audio, video, or images to depict fabricated events or scenarios. While initially used for entertainment purposes, deepfakes pose significant risks in various domains, including politics, cybersecurity, and social media. With the proliferation of deepfake technology, the potential for misinformation and fraudulent activities has become a significant concern. In response, numerous research efforts have been devoted to developing effective deepfake detection methods. This paper presents a comprehensive review of deepfake detection techniques utilizing machine learning algorithms over several datasets. We analyze and compare a diverse range of Machine-Learning approaches. By synthesizing insights from a wide array of research contributions, this review aims to provide a comprehensive understanding of the current state of deepfake detection using machine learning and to guide future research directions in mitigating the threats posed by deepfakes. Despite significant advancements, current state-of-the-art approaches exhibit limitations, such as sample size dependence, misclassification risks, reliance on specific data types, or inefficiency in handling complex architectures. To address these limitations and enhance detection accuracy, future research could explore the integration and fusion of multiple detection models.