Explainable Deep Learning Approach for Detecting Money Laundering Transactions in Banking System

Money laundering has been a global issue since last few decades. The continuous change in technology, fraud patterns and regulatory landscape is making it even harder. To assist the compliance officers with investigation, Artificial Intelligence (AI) based decision support systems are developed which suffers from a high false negative rate of alerts without having much of transparency in the predictions. This thesis presents a study on the application of Convolutional Neural Network (CNN) method for predicting the suspicious money laundering transactions and explaining the predictions using SHapley Additive exPlanations (SHAP) XAI method. The results showed that the CNN model outperformed other models, indicating better handling of compliance risk. On the contrary, CNN model showed higher number of false positives compared to other models which indicates more operational efforts. From the Banks perspective, reducing the compliance risk is more important than operational efforts. Hence f_β score was calculated using β=3 to measure the performance. The scores were CNN - 78.23%, XGB – 62.09%, RF – 61.09%, and SVM - 30.86%. The SHAP XAI method applied on CNN model to interpret the predictions made by CNN model, could identify the key features influencing the predictions.

Author dc:creator

• Kute, Dattatray Vishnu

Statement dc:rights

• info:eu-repo/semantics/openAccess
• The author owns the copyright in this thesis including all reproduction and reuse rights for the work. The work may not be altered without the permission of the copyright owner. Attribution is essential when quoting or paraphrasing from this thesis.
• © 2022 Dattatray Vishnu Kute
• au.edu.uts.lib/ppc

Language dc:language.iso

en_US

Handle dc:identifier.uri

http://hdl.handle.net/10453/167740

OAI identifier oai:identifier

oai:opus.lib.uts.edu.au:10453/167740