An Advanced Deepfake Detection System leveraging DeepfakeBench and Explainable AI
DFRWS Digital Forensics Conference
Saakshi Gupta
// ABSTRACT
Research poster presented at the DFRWS Digital Forensics Conference on advanced deepfake detection methodologies. Combines the DeepfakeBench evaluation framework with explainable-AI techniques to make detection performance both measurable and interpretable.
Summary
Poster presentation at the DFRWS (Digital Forensics Research) Conference, covering an advanced deepfake detection system that combines two pieces of prior work:
- DeepfakeBench — a standardised evaluation framework for deepfake detection models, enabling apples-to-apples comparison across detectors
- Explainable AI — SHAP and LIME-based saliency to surface why a detection fired
Why combine them
DeepfakeBench tells you whether your detector works. Explainable AI tells you why. Combining them lets practitioners pick the right detector for a given scenario and defend its conclusions to a downstream investigator.
Audience
The DFRWS audience is primarily digital forensics practitioners — the people who actually have to use these tools in court-facing investigations. Designing the system around their needs (explanation-first, benchmarked, reproducible) was the explicit goal.