Deepfake Video Detection using Machine Learning

Abstract - The rise of DeepFake (DF) technology poses a significant threat to the authenticity and reliability of digital media. While such videos once required expert skills and high-end software, the recent advancement of deep learning tools has made the generation of hyper-realistic synthetic media widely accessible. This paper presents a robust and efficient deep learning-based solution for detecting DeepFake videos by combining spatial and temporal analysis using a Convolutional Neural Network (CNN) and a Long Short-Term Memory (LSTM) based Recurrent Neural Network (RNN). Our final system leverages a ResNeXt50 CNN for frame-level feature extraction and an LSTM layer to model temporal inconsistencies across sequential video frames. Extensive experiments conducted on a balanced dataset—comprising both real and manipulated videos sourced from FaceForensics++, YouTube, and the DeepFake Detection Challenge dataset—demonstrate the high accuracy and reliability of our model. Additionally, the developed solution is deployed as a web-based platform allowing real-time DeepFake video detection with user-friendly interaction. This paper outlines the design, implementation, results, and scope of our complete DeepFake detection system. Keywords: DeepFake Detection, CNN, LSTM, ResNeXt, Temporal Analysis, Video Forensics, Media Authenticity, Deep Learning, Media Verification, Face Forensics