DeepGuard: A Hybrid CNN-LSTM Framework for Robust Deepfake Video Detection with Spatiotemporal Analysis

Deepfake technology has developed very quickly, allowing highly realistic artificial media to be created that causes serious risks to privacy, security, and disinformation. This paper introduces a deep learning method for identifying deepfake videos by applying a hybrid Convolutional Neural Network (CNN) and Long Short-Term Memory (LSTM) model. The proposed model takes advantage of spatial information learned by a CNN (Xception backbone) and temporal relationships learned by an LSTM in order to label videos as real or fake. The model was trained on the FaceForensics++ dataset and evaluated on the Celeb-DF dataset with an accuracy of 82%. A web application was also implemented for real-time deepfake detection. The outcome proves the efficacy of the suggested architecture in detecting manipulated videos with challenges and future enhancements