Enhancing Deepfake Video Detection Performance with a Hybrid CNN Deep Learning Model

This These days, a lot of phoney photos and movies are produced using modern AI (Artificial Intelligence) technology and software, which occasionally leaves traces of manipulation. Videos may be used in numerous unethical ways to intimidate, quarrel, or incite fear in others. Making ensuring that these techniques aren't applied to produce phoney videos is crucial. Deep Fake is an AI-based method for creating synthetic photographs of people. They are made by merging and overlaying preexisting videos with the original videos. This research develops a system that extracts frame-level characteristics using Convolutional Neural Network (CNN) hybrid made out of Xception and InceptionResnet v2. We do an experimental investigation with Kaggle's DFDC deep fake detecting challenge. By utilizing this dataset's for testing and training, these deep learning-based techniques are tuned to improve accuracy while cutting down on training time. Our results showed a 0.985 accuracy, 0.96 recall, 0.98 f1-score, and 0.968 support.