An automated approach to financial fraud management
An automated approach to financial fraud management
Financial fraud is a constantly evolving problem, projected to reach $40 Billion in global losses by 2027. The operational cost of managing fraud systems is high and continues to increase, requiring fraud teams to expand as the volume of payments and fraud loss increase. This major problem is rarely identified. The aim of this thesis is to develop methods for fraud detection process automation, enabling the fraud teams to deal with the ever increasing payment and fraud volumes efficiently. A novel method, AutoPilotML is proposed, which develops optimal ensembles of machine-generated and manual fraud rules to detect as much fraud as possible with as little human intervention as possible. AutoPilotML is shown to drastically reduce the manual effort involved in managing fraud strategies, improving operational efficiency.
The time delay between a payment being made and becoming marked as fraud/genuine can take up to as much as 3 months, leaving the most recent novel fraud behaviour unchallenged. Two unsupervised outlier detection methods are proposed and evaluated to show how they can be used for detecting novel fraud trends.
The combination of AutoPilotML and one of the outlier detection models is studied to understand how the systems can be combined to create a fully automated fraud detection system.
University of Southampton
Tearle, Oliver John
cbc80967-ddab-493b-ac25-0ed4ea3d3fa8
12 December 2025
Tearle, Oliver John
cbc80967-ddab-493b-ac25-0ed4ea3d3fa8
Frey, Jeremy
ba60c559-c4af-44f1-87e6-ce69819bf23f
Brodzki, Jacek
b1fe25fd-5451-4fd0-b24b-c59b75710543
Niranjan, Mahesan
5cbaeea8-7288-4b55-a89c-c43d212ddd4f
Tearle, Oliver John
(2025)
An automated approach to financial fraud management.
University of Southampton, Doctoral Thesis, 253pp.
Record type:
Thesis
(Doctoral)
Abstract
Financial fraud is a constantly evolving problem, projected to reach $40 Billion in global losses by 2027. The operational cost of managing fraud systems is high and continues to increase, requiring fraud teams to expand as the volume of payments and fraud loss increase. This major problem is rarely identified. The aim of this thesis is to develop methods for fraud detection process automation, enabling the fraud teams to deal with the ever increasing payment and fraud volumes efficiently. A novel method, AutoPilotML is proposed, which develops optimal ensembles of machine-generated and manual fraud rules to detect as much fraud as possible with as little human intervention as possible. AutoPilotML is shown to drastically reduce the manual effort involved in managing fraud strategies, improving operational efficiency.
The time delay between a payment being made and becoming marked as fraud/genuine can take up to as much as 3 months, leaving the most recent novel fraud behaviour unchallenged. Two unsupervised outlier detection methods are proposed and evaluated to show how they can be used for detecting novel fraud trends.
The combination of AutoPilotML and one of the outlier detection models is studied to understand how the systems can be combined to create a fully automated fraud detection system.
Text
Oliver Tearle Thesis FINAL PDFA
- Version of Record
Restricted to Repository staff only until 12 December 2028.
Text
Final-thesis-submission-Examination-Mr-Oliver-Tearle
Restricted to Repository staff only
More information
Published date: 12 December 2025
Identifiers
Local EPrints ID: 507766
URI: http://eprints.soton.ac.uk/id/eprint/507766
PURE UUID: 707d5dfe-2508-494e-9040-26ecd3dec028
Catalogue record
Date deposited: 06 Jan 2026 12:58
Last modified: 08 Jan 2026 02:58
Export record
Contributors
Author:
Oliver John Tearle
Thesis advisor:
Mahesan Niranjan
Download statistics
Downloads from ePrints over the past year. Other digital versions may also be available to download e.g. from the publisher's website.
View more statistics