A Comprehensive Survey on Explainable Deepfake Detection: Techniques, Challenges, and Future Directions

Detecting deepfakes is getting harder as more sophisticated deep learning-based models are used to create deepfakes. XAI is a transparent and interpretable solution to the deepfake detection model as an effective methodology to tackle these challenges. This paper aims to demonstrate how various XAI methods (advanced visualization methods and feature interpretation methods) can be used to help understand how these models function in identifying and classifying deepfakes. The integrative XAI with traditional AI-based methods has also shown the potential in the detection of various categories of deepfakes, including identity swap-based deepfakes and image-based deep learning-based deepfakes. However, in both these systems there is a great need to enhance the robustness and trustworthiness of these systems with significant improvements in the interpretability of AI models through new XAI techniques that can extend the application of deepfake detection across domains such as social media authentication, digital evidence, and automated content control.