FRAUD DETECTION USING DATA-DRIVEN APPROACH

Session

Computer Science and Communication Engineering

Description

The extensive use of internet is continuously drifting businesses to incorporate their services in the online environment. One of the first spectrums to embrace this evolution was the banking sector. In fact, the first known online banking service came in 1980. It was deployed from a community bank located in Knoxville, called the United American Bank. Since then, internet banking has been offering ease and efficiency to costumers in completing their daily banking tasks. The ever increasing use of internet banking and the large number of online transactions, increased fraudulent behaviour also. As if fraud increase wasn’t enough, the massive number of online transactions further increased the data complexity. Modern data sources are not only complex but generated at high speed and in real time as well. This presents a serious problem and a definite reason why more advanced solutions are desired to protect financial service companies and credit card holders. Therefore, this thesis aims to construct an efficient fraud detection model which is adaptive to costumer behaviour changes and tends to decrease the fraud manipulation, by detecting and filtering fraud in real-time. In order to achieve this aim, a review of various methods is conducted, adding above a personal experience working at a Banking sector, specifically in Fraud Detection office. Unlike the majority of reviewed methods, the proposed model in this thesis is able to detect fraud in the moment of occurrence using an incremental classifier. The evaluation on synthetic data, based on fraud scenarios selected in collaboration with domain experts that replicate typical, real-world attacks, shows that this approach correctly ranks complex frauds. In particular, our proposal detects fraudulent behaviour and anomalies with up to 97% detection rate while maintaining a satisfying low cost.

Keywords:

Banking · Online · Data · Fraud

Session Chair

Bertan Karahoda

Session Co-Chair

Besnik Qehaja

Proceedings Editor

Edmond Hajrizi

ISBN

978-9951-437-96-7

First Page

96

Last Page

101

Location

Lipjan, Kosovo

Start Date

31-10-2020 10:45 AM

End Date

31-10-2020 12:30 PM

DOI

10.33107/ubt-ic.2020.502

This document is currently not available here.

Share

COinS
 
Oct 31st, 10:45 AM Oct 31st, 12:30 PM

FRAUD DETECTION USING DATA-DRIVEN APPROACH

Lipjan, Kosovo

The extensive use of internet is continuously drifting businesses to incorporate their services in the online environment. One of the first spectrums to embrace this evolution was the banking sector. In fact, the first known online banking service came in 1980. It was deployed from a community bank located in Knoxville, called the United American Bank. Since then, internet banking has been offering ease and efficiency to costumers in completing their daily banking tasks. The ever increasing use of internet banking and the large number of online transactions, increased fraudulent behaviour also. As if fraud increase wasn’t enough, the massive number of online transactions further increased the data complexity. Modern data sources are not only complex but generated at high speed and in real time as well. This presents a serious problem and a definite reason why more advanced solutions are desired to protect financial service companies and credit card holders. Therefore, this thesis aims to construct an efficient fraud detection model which is adaptive to costumer behaviour changes and tends to decrease the fraud manipulation, by detecting and filtering fraud in real-time. In order to achieve this aim, a review of various methods is conducted, adding above a personal experience working at a Banking sector, specifically in Fraud Detection office. Unlike the majority of reviewed methods, the proposed model in this thesis is able to detect fraud in the moment of occurrence using an incremental classifier. The evaluation on synthetic data, based on fraud scenarios selected in collaboration with domain experts that replicate typical, real-world attacks, shows that this approach correctly ranks complex frauds. In particular, our proposal detects fraudulent behaviour and anomalies with up to 97% detection rate while maintaining a satisfying low cost.