A Comparative Analysis and Study of a Fast Parallel CNN Based Deepfake Video Detection Model with Feature Selection (FPC-DFM)

Deep learning is an efficient and practical method that has been widely applied in numerous fields. Videos created with swapped faces in a video, altered facial emotions, changed genders, fraudulent video content generation, and altered facial features are referred to as “DeepFake” videos. These videos are created utilizing deep learning technology called generative adversarial networks. Fake videos are used to stir up political agitation, commit acts of terrorism, and demand money. A fast Parallel CNN-based deepfake video detection model with feature selection is the new model we presented in this project (FPC-DFM). In order to identify Deepfake videos, the FPC DFM architecture uses feature extraction, feature selection, and prediction. The FPC-DFM model extracted features using three convolutional models: EfficientN et, VGG16, and ResNet as well as Principal Component Analysis (PCA)-based feature selection and Support Vector Machine-based classification. We offer a new architecture for capturing the video frame features that will be utilized to determine if the video is real or fake by utilizing deep learning techniques. In comparison to other pre-trained models like VGG16-TL, EfficientNet-TL and Resnet50-TL and our results demonstrated that FPC-DFM has the best performance and the highest accuracy of 96.50%.