Abstract: Rapid advancements in AI, machine learning, and deep learning over the past few decades have led to the development of new methods and tools for altering multimedia. Despite the facts that technology has primarily been utilized for good reasons, including entertainment and education, unscrupulous people have nonetheless taken advantage of it for illegal or sinister ends. For instance, realistic-seeming, high-quality phony films, pictures, or sounds have been produced with the intention of propagandizing false information, inciting hatred and political unrest, or even harassing and blackmailing individuals. Recently, the highly-reproduced, lifelike, and altered videos have come to be known as Deepfake. Since then, a number of strategies have been detailed in the literature to address the issues brought up by Deepfake. By safeguarding data, identifying deepfakes, and preventing media manipulation, deepfake video detection contributes to cybersecurity. Videos that are the original data are altered for a number of reasons. It's critical to be able to spot this kind of misleading information. In the social media age, identity theft is seen as the main issue. In order to explore the most promising new methods for deepfake video detection, this paper examines the most recent research findings from the community. This system uses convolutional neural networks (CNNs) and long term memory (LSTM) to distinguish between real and fake video frames. This also involves the application of the Densenet algorithm, XGBoosting classifier, and YOLO Face detector. Faces in videos can be found using the YOLO face detector. To help detect visual artifacts in the video frames, InceptionResNetV2 CNN is used to extract discriminant spatial features of these faces. The XGBoost classifier uses these visual characteristics to assist differentiate between real and deepfake films.

Keywords: Fake video, YOLO, CNN, deep learning.

| DOI: 10.17148/IARJSET.2024.11495