Deepfake Video Detection Using Machine Learning

Deepfake Technology, powered by advancements in generative models such as Generative Adversarial Networks (GANs), has made it increasingly easy to manipulate video content with a high degree of realism. While this technology has legitimate applications in entertainment and accessibility, it also poses serious threats to individual privacy, public trust, and national security when used maliciously. This paper addresses the growing challenge of detecting deepfake videos by proposing a hybrid deep learning-based detection system that combines the strengths of both Convolutional Neural Networks (CNNs) and Recurrent Neural Networks (RNNs). The proposed model utilizes the ResNeXt50 architecture to extract robust spatial features from individual video frames and employs Long Short-Term Memory (LSTM) networks to analyze temporal dependencies between frames. This dual approach allows the system to identify inconsistencies not just in individual frames but also in the sequence of actions and expressions across the video The model is trained on publicly available deepfake datasets and demonstrates a high level of accuracy, precision, and recall in classifying real versus manipulated videos. Furthermore, the model is integrated into a user-friendly web application using the Flask framework, The results indicate that this hybrid CNN-RNN architecture outperforms many traditional detection approaches, making it a promising tool for video authentication in real-world scenarios. The proposed system contributes to ongoing efforts in digital forensics and deepfake mitigation by offering a scalable, accurate, and efficient solution.