DeepFake Reporter Guard: A Deep Learning and Explainability-based System for Reliable News Verification

Abstract

Modern machine learning methods such as Generative Adversarial Networks (GANs) are used to automatically produce the fake video content that looks completely real. This threatens the integrity of digital media, journalism, and the public trust. In response to these threats, the DeepFake Reporter Guard offers an explainable AI-based web platform for the journalists to verify the authenticity of a news video. For regional specificity, this system uses XceptionNet model, which was trained on the INDIFACE dataset consisting of Indian faces. The detection methodology involves measuring the facial inconsistencies across the frame level. Once deepfake videos are identified, Grad-CAM-based explainability highlights the manipulated frames in the video and provides visual explanation. The explanation is presented in a way that non-technical users can understand it. The user-friendly web interface allows the journalists or users to upload the videos and get the instant authenticity reports telling whether the video is real or fake with a clear justification. The proposed method is robust and adaptable, making it suitable for real world situations.