BELFAL: A Blockchain-based Ensemble Learning Framework for Anti-money Laundering in Crypto-Currency Markets

Crypto-Currency has witnessed a growing trend of adoption and skyrocketing market size in recent years, imposing significant impacts and challenges to financial industries. One of the challenges that attract intensive interests from academia and industries is money laundering with crypto-currency, due to such regulation-resistant features as decentralization, anonymity and faster trading speed. The existing works propose to address this anti-money laundering (AML) issue by optimizing an individual machine learning algorithm or simply ensembling homogeneous models with fixed rules, and thus may fail in adapting to various AML scenarios in dynamic and open crypto-currency markets. In this paper, we aim to tackle this problem by proposing a novel blockchain-based ensemble learning framework for anti-money laundering(BELFAL) in crypto-currency markets. BELFAL can improve prediction performance of AML tasks through parallel, collaborative ensembling of multiple heterogenous models, and leverage blockchain and smart contracts for flexible scheduling of ensemble rules. We establish a preliminary prototype based on Ethereum and Inter-Planetary File System (IPFS) to evaluate our framework, and conduct experiments on Elliptic, an opensource dataset of Bitcoin transactions. The results verify that our ensemble model and framework outperform individual models, and can help realize flexible scheduling of models.