Authors:  Arvind Mithal∗, Ethan Song and Daniel Thompson

JSA-Vol. 4 (2025),

1 MIT-IBM Watson AI Lab, Cambridge, MA, United States.

* Correspondence: arvind@csail.mit.edu

Received: 2 November 2024; Accepted: 5 April 2025; Published: 20 April 2025

Abstract: Money laundering detection in large-scale financial transaction networks remains a critical and challenging task due to the scale of data, evolving criminal strategies, and strict regulatory requirements for interpretability. Recent advances have demonstrated that sender–receiver abstractions provide an effective and scalable alternative to explicit subgraph enumeration by representing suspicious activity through boundary node sets. However, existing approaches remain largely static and offer limited support for temporal reasoning and forensic explanation. In this paper, we propose a Temporal and Explainable Sender–Receiver Framework that extends sender–receiver modeling by integrating temporal dynamics, transaction ordering, and value-flow behavior. We further introduce an explanation layer that generates time-respecting evidence paths and interpretable laundering indicators to support human investigators. The proposed framework preserves scalability while significantly improving detection accuracy and interpretability. Extensive experimental analysis demonstrates the effectiveness of the proposed approach in realistic large-scale transaction networks.

Keywords: Anti-money laundering; transaction graphs; temporal networks; explainable AI; graph learning