Evaluating Models for Deepfake Detection: A Comparative Study

Deepfake detection poses a significant challenge in digital forensics due to increasingly advanced AI-generated videos. This study evaluates three models using the FaceForensics++ and DeeperForensics-1.0 datasets. The preprocessing involved video encoding, renaming, trimming, frame extraction, face detection, and data loading. The first model, a Convolutional Neural Network (CNN), achieved 84% accuracy. The second model, Xception, an efficient CNN with depth-wise separable convolutions and residual connections, attained 89% accuracy. The third model, combining CNN with a Recurrent Neural Network (RNN) and LSTM layers, significantly improved detection accuracy to 97%. This hybrid model highlights the importance of capturing spatial and temporal features in deepfake detection, demonstrating the efficacy of advanced deep learning techniques in addressing the deepfake threat.