Preparing Blockchain Transaction Data for AI- Based AML Detection: A Reproducible Cloud Pipeline

Abstract

Public blockchains provide full transaction histories, but they are not ready for anti-money laundering (AML) modelling in raw form. The data are large, chain-specific, and sensitive to reorganisation near the chain tip. This paper presents a reproducible cloud pipeline for preparing Bitcoin and Ethereum transaction data for AI-based AML detection. The design uses AWS Public Blockchain Data for historical backfill, controlled snapshot versioning in private Amazon S3, and incremental tailing through Bitcoin Core and Ethereum JSON-RPC. Raw partitions are transformed into canonical blocks, transactions, value flows, and Ethereum event views. Provenance fields are recorded at each stage so that datasets can be rebuilt and audited. The paper also defines a chain-aware finality policy, using confirmation depth for Bitcoin and safe or finalized anchors for Ethereum. The result is a practical blueprint for building AI-ready blockchain datasets that are repeatable, inspectable, and suitable for later model training and compliance review.