Enhancing Deepfake Detection: A Comparative Study of MesoNet, XceptionNet, EfficientNet, Optical flow and XceptionNet + Optical flow

Deepfake videos, which are manipulated to show false information, are becoming a major challenge in digital media. In this study, we explore four techniques—MesoNet, XceptionNet, Optical Flow, EfficientNet and an Ensemble Hybrid Model to detect deepfake videos. MesoNet is a lightweight neural network that focuses on detecting tampered images; XceptionNet is a more advanced model that identifies subtle fake features in videos, Optical Flow analyzes the movement between video frames to find inconsistencies that may indicate tampering, EfficientNet provides a balance between computational efficiency and accuracy by scaling neural networks effectively. We tested these models individually and in combination to determine the most effective approach for deepfake detection. By leveraging both spatial features (visual details) and temporal features (movement across frames), we achieved better identification of fake videos. This study highlights the importance of integrating spatial and temporal features along with different techniques to address growing threat of deepfakes and ensure the integrity of digital content in real-world applications.