DEEP FAKE DETECTION USING MACHINE LEARNING

Abstract

The recent proliferation of free, deep learning-based tools has democratized the creation of high-fidelity “deepfake” videos, where facial exchanges exhibit minimal manipulation artifacts. While advancements in visual effects have facilitated video manipulation for decades, deep learning has ushered in an era of unprecedented realism and accessibility for generating such “AI-synthesized media.” While crafting deepfakes is now relatively straightforward, their detection remains a significant challenge due to the complexities involved in training algorithms to recognize these subtle manipulations. This paper presents a novel deepfake detection method employing a combined convolutional neural network (CNN) and recurrent neural network (RNN) architecture. The CNN extracts frame-level features, and these features are subsequently fed into an RNN trained to classify videos as manipulated or authentic. The RNN’s ability to learn temporal dependencies allows it to effectively detect inconsistencies between frames, often introduced by deepfake creation tools. We evaluate our system on a comprehensive dataset of synthetic videos and demonstrate competitive performance, highlighting the potential of our relatively simple architecture in addressing this critical challenge.