Logo image
Back
A Decentralized Approach to Deepfake Detection Using Blockchain-Based Federated Learning in Forensic Contexts
Conference proceeding   Peer reviewed

A Decentralized Approach to Deepfake Detection Using Blockchain-Based Federated Learning in Forensic Contexts

Maryam Taeb, Shonda Bernadin and Hongmei Chi
IEEE International Conference on Big Data, pp.4416-4424
International Conference on Big Data (BigData) (Macau, China, 12/08/2025–12/11/2025)
12/08/2025
DOI: https://doi.org/10.1109/BigData66926.2025.11402187

Metrics

1 Record Views

Abstract

In today's digital landscape, smartphones serve as vital tools for documenting life events, often providing crucial evidence in legal proceedings. However, the widespread distribution of digital evidence across devices and social platforms presents challenges to law enforcement agencies. These challenges are compounded by the complexity of extracting and preserving digital media while maintaining its integrity. The rise of AI-generated editing tools and synthetic data has introduced new risks to digital forensics. Deepfake technology enables sophisticated manipulation of digital content, threatening evidence authenticity and facilitating fraud, deception, and harassment. To address these challenges, we present BFDD (Blockchain-Based Decentralized Federated Learning Framework for Deepfake Detection in Digital Forensic Scenarios), a comprehensive solution designed for law enforcement. BFDD integrates blockchain and federated learning to: (1) streamline digital evidence verification and (2) enhance AI-generated synthetic media detection using advanced computer vision and large language models. BFDD achieves 97% accuracy in detecting deepfake images and 99% accuracy in identifying fake news. It also ensures efficiency and cost- effectiveness, reaching 40 TPS for workloads between 500 KB to 800 KB, while maintaining a 30-second latency for larger datasets. The framework reduces transaction costs by 99.94%, with an average fee of 1.14 per transaction, compared to Ethereum's 1943.35. Furthermore, its hash-based storage mechanism optimizes space, reducing storage requirements by 127.6 times compared to storing original evidence, making it a scalable, secure, and cost-efficient solution for digital forensic applications.

Details

Logo image