Qualifying and raising anti-money laundering alarms with deep learning

Rasmus Ingemann Tuffveson Jensen; Alexandros Iosifidis

doi:10.1016/j.eswa.2022.119037

Qualifying and raising anti-money laundering alarms with deep learning

Rasmus Ingemann Tuffveson Jensen^*, Alexandros Iosifidis

^*Corresponding author for this work

Department of Electrical and Computer Engineering - Signal Processing and Machine learning

Research output: Contribution to journal/Conference contribution in journal/Contribution to newspaper › Journal article › Research › peer-review

Abstract

We propose a deep learning approach to qualify and raise anti-money laundering alarms in banks. The motivating idea is to replace predefined rules with latent features automatically extracted from transaction sequences. To this end, we experiment with recurrent and Transformer encoder layers. We test the approach with a large data set from Spar Nord Bank. Our best model employs gated recurrent units and self-attention. When used to qualify alarms raised by a traditional AML system, the model is able to reduce the number of false positives by more than 33.3% while retaining 98.8% of all true positives. In a small experiment, we also use the model to raise 75 new alarms on clients with high risk scores but with no traditional alarm inquiries in the last 12 months. After expert review, this prompted 26 clients to be reported to Danish authorities.

Original language	English
Article number	119037
Journal	Expert Systems with Applications
Volume	214
ISSN	0957-4174
DOIs	https://doi.org/10.1016/j.eswa.2022.119037
Publication status	Published - Mar 2023

Keywords

Anti-money laundering
Deep learning
Money laundering

Access to Document

10.1016/j.eswa.2022.119037

Cite this

@article{e9ee4f793b3843eeb467c2b1bec1868d,

title = "Qualifying and raising anti-money laundering alarms with deep learning",

abstract = "We propose a deep learning approach to qualify and raise anti-money laundering alarms in banks. The motivating idea is to replace predefined rules with latent features automatically extracted from transaction sequences. To this end, we experiment with recurrent and Transformer encoder layers. We test the approach with a large data set from Spar Nord Bank. Our best model employs gated recurrent units and self-attention. When used to qualify alarms raised by a traditional AML system, the model is able to reduce the number of false positives by more than 33.3% while retaining 98.8% of all true positives. In a small experiment, we also use the model to raise 75 new alarms on clients with high risk scores but with no traditional alarm inquiries in the last 12 months. After expert review, this prompted 26 clients to be reported to Danish authorities.",

keywords = "Anti-money laundering, Deep learning, Money laundering",

author = "Jensen, {Rasmus Ingemann Tuffveson} and Alexandros Iosifidis",

note = "Publisher Copyright: {\textcopyright} 2022 Elsevier Ltd",

year = "2023",

month = mar,

doi = "10.1016/j.eswa.2022.119037",

language = "English",

volume = "214",

journal = "Expert Systems with Applications",

issn = "0957-4174",

publisher = "Pergamon Press",

}

TY - JOUR

T1 - Qualifying and raising anti-money laundering alarms with deep learning

AU - Jensen, Rasmus Ingemann Tuffveson

AU - Iosifidis, Alexandros

PY - 2023/3

Y1 - 2023/3

N2 - We propose a deep learning approach to qualify and raise anti-money laundering alarms in banks. The motivating idea is to replace predefined rules with latent features automatically extracted from transaction sequences. To this end, we experiment with recurrent and Transformer encoder layers. We test the approach with a large data set from Spar Nord Bank. Our best model employs gated recurrent units and self-attention. When used to qualify alarms raised by a traditional AML system, the model is able to reduce the number of false positives by more than 33.3% while retaining 98.8% of all true positives. In a small experiment, we also use the model to raise 75 new alarms on clients with high risk scores but with no traditional alarm inquiries in the last 12 months. After expert review, this prompted 26 clients to be reported to Danish authorities.

AB - We propose a deep learning approach to qualify and raise anti-money laundering alarms in banks. The motivating idea is to replace predefined rules with latent features automatically extracted from transaction sequences. To this end, we experiment with recurrent and Transformer encoder layers. We test the approach with a large data set from Spar Nord Bank. Our best model employs gated recurrent units and self-attention. When used to qualify alarms raised by a traditional AML system, the model is able to reduce the number of false positives by more than 33.3% while retaining 98.8% of all true positives. In a small experiment, we also use the model to raise 75 new alarms on clients with high risk scores but with no traditional alarm inquiries in the last 12 months. After expert review, this prompted 26 clients to be reported to Danish authorities.

KW - Anti-money laundering

KW - Deep learning

KW - Money laundering

UR - http://www.scopus.com/inward/record.url?scp=85141293542&partnerID=8YFLogxK

U2 - 10.1016/j.eswa.2022.119037

DO - 10.1016/j.eswa.2022.119037

M3 - Journal article

AN - SCOPUS:85141293542

SN - 0957-4174

VL - 214

JO - Expert Systems with Applications

JF - Expert Systems with Applications

M1 - 119037

ER -

Qualifying and raising anti-money laundering alarms with deep learning

Abstract

Keywords

Access to Document

Other files and links

Fingerprint

Cite this