Anomaly detection in cross-country money transfer temporal networks
This paper explores anomaly detection through temporal network analysis. Unlike many conventional methods, relying on rule-based algorithms or general machine learning approaches, our methodology leverages the evolving structure and relationships within temporal networks, that can be used to model f...
Gespeichert in:
Hauptverfasser: | , , , , , , |
---|---|
Format: | Artikel |
Sprache: | eng |
Schlagworte: | |
Online-Zugang: | Volltext bestellen |
Tags: |
Tag hinzufügen
Keine Tags, Fügen Sie den ersten Tag hinzu!
|
Zusammenfassung: | This paper explores anomaly detection through temporal network analysis.
Unlike many conventional methods, relying on rule-based algorithms or general
machine learning approaches, our methodology leverages the evolving structure
and relationships within temporal networks, that can be used to model financial
transactions. Focusing on minimal changes in stable ecosystems, such as those
found in large international financial institutions, our approach utilizes
network centrality measures to gain insights into individual nodes. By
monitoring the temporal evolution of centrality-based node rankings, our method
effectively identifies abrupt shifts in the roles of specific nodes, prompting
further investigation by domain experts.
To demonstrate its efficacy, our methodology is applied in the Anti-Financial
Crime (AFC) domain, analyzing a substantial financial dataset comprising over
80 million cross-country wire transfers. The goal is to pinpoint outliers
potentially involved in malicious activities, aligning with financial
regulations. This approach serves as an initial stride towards automating AFC
and Anti-Money Laundering (AML) processes, providing AFC officers with a
comprehensive top-down view to enhance their efforts. It overcomes many
limitations of current prevalent paradigms, offering a holistic interpretation
of the financial data landscape and addressing potential blindness to phenomena
that cannot be effectively estimated through single-node or narrowly focused
transactional approaches. |
---|---|
DOI: | 10.48550/arxiv.2311.14778 |