Deepfake Detection using XAI based Deep Fusion Models
A. Anitha, Lis Maria Saju, Balakrishnan Kamaraj · 2025
Deepfake technology has rapidly evolved, posing a serious threat to the authenticity of digital media and contributing to the spread of misinformation. The manipulation of media content raises significant concerns across sectors like politics, entertainment, and social media, undermining public trust in digital information. In response, this research proposes an advanced deepfake detection model that integrates Explainable Artificial Intelligence (XAI) techniques within a hybrid deep learning architecture. The proposed model combines ResNet50 and Vision Transformer (ViT) to capture spatial and contextual features effectively. While exploring additional methods such as facial landmark detection and frequency domain analysis, the final model focuses on the ResNet50-ViT combination for its superior performance to fusion model. To ensure transparency, the model employs XAI techniques such as LIME (Local Interpretable Model-agnostic Explanations) and SHAP (SHapley Additive exPlanations), offering interpretability by revealing the key factors that influence its predictions. This dual focus on high detection accuracy and interpretability ensures that the model not only detects deepfakes effectively but also builds trust in automated decision-making processes. By leveraging cutting-edge deep learning and XAI methods, this framework offers a more reliable, transparent, and adaptable solution for preserving digital media authenticity across various applications, ultimately contributing to combating misinformation and fostering trust in digital content.