DEEPFAKE IDENTIFICATION USING MACHINE LEARNING TOOLS

Abstract

Deepfake technology brings new ideas and dangers to the digital world, thanks to quick advances in AI and generative models. The fake videos and pictures do a better job of showing how people really act, which is a serious threat to authenticity, privacy, and public trust. In this paper, we examine deepfakes through the lenses of social implications and technical detection by introducing a comprehensive detection framework that incorporates various deep learning models: Hybrid (ResNet-50 + DCT Fusion), CNN, LSTM, Transformer, and Spectral Analysis architectures. We used the Kaggle "1000-videos-split" dataset to train and test the proposed system. This dataset has a good mix of real and fake media. The CNN and Hybrid detectors had the highest accuracy in experiments, finding the best results in different test cases with 99.2% accuracy. Beyond the technical results, this investigation also underlined an integral approach with AI detection, ethics, and design practices, accompanied by a call for public awareness in preserving authenticity in digital communication.