This article first appeared in Data Science Briefings, the DataMiningApps newsletter. Subscribe now for free if you want to be the first to receive our feature articles, or follow us @DataMiningApps. Do you also wish to contribute to Data Science Briefings? Shoot us an e-mail over at firstname.lastname@example.org and let’s get in touch!
Analytics in the battle against fraud
BNP Paribas Fortis and the KU Leuven have recently joined forces in a research program to combat transactional fraud through analytics. As a socially responsible enterprise, BNP Paribas Fortis financially supports the KU Leuven Fraud Analytics chair to stimulate co-creation and to invest in fraud analytics expertise. In light of the chair, the Bank and the University organized a half-day seminar on Transactional Fraud Analytics on the 6th of June 2017.
The event was intended to hear what bank experts and academic researchers have to share about the role and importance of analytics in the banking industry, transactional fraud from an analytics point of view, and the latest findings and methods in this domain. The seminar also gave the opportunity to qualified graduates, peer practitioners and stakeholders to exchange their thoughts and opinions in this exciting field during Q&As and networking sessions.
The kick-off of the seminar was given by Jo Coutuer, Chief Data Officer of BNP Paribas Fortis, who gave a keynote on the role and importance of analytics in the bank. Many activities of the Bank nowadays depend on data, including fraud detection. Gathering, maintaining and analyzing this data is not a one-man task, but requires a team of specialists with different expertise that need to work together seamlessly. Their activities are supported by powerful tooling.
“Fraud matters” says Wim Bartsoen, head of Information Security Governance, Norms, Monitoring and Control at BNP Paribas Fortis. Successful fraud management minimizes potential losses for both bank and customer and keeps friction limited. Its purpose is to preserve people’s faith in online channels and to maintain the Bank’s public image of trusted institution in the face of ever more creative ways through which criminals try to steal money.
Analytics point of view
Transactional fraud includes a type of fraud that is committed via credit transfers. Fraudsters try to steal money and personal credentials through different modus operandi (MO’s) like phishing or hacking. Alejandro Marcos Alvarez, data scientist at BNP Paribas Fortis, speaks about the specific nature of transactional fraud from an analytics point of view. The data that is available within the Bank is diverse and can be of a financial, transactional and personal nature or even related to the online device used by the customer. Besides the type also the amount of data is an important aspect. The Bank processes very large volumes of data, around 1 million transfers each day. Fraud is typically a rare event (less than 1 fraud case in 10 million transactions), so the biggest challenge for a data scientist is to detect fraudulent transfers within these large stacks of data. Moreover, the incoming data must be processed in (near) real-time so the potential fraudulent nature of a transfer can be determined in less than a few seconds. Analytics is used to model different aspects of a credit transfer like the session of a user and the behavior of a customer’s account. Mr. Marcos adds that fraud detection techniques must be able to adapt quickly since fraudsters are agile; existing MO’s are adapted and new ones are created.
A key aspect of a transaction is the beneficiary. Sebastiaan Höppner, PhD researcher at the KU Leuven, focuses on a particular kind of beneficiaries called money mules. These are intermediaries who make their bank account available (either unwittingly or maliciously) to criminals in order to transfer stolen money, often from
online fraud like phishing and malware scams. Money mules transfer the stolen proceeds on behalf of others and are usually rewarded by a small part of the stolen money. Mr. Höppner works as data scientist at BNP Paribas Fortis in the context of the Fraud Analytics chair. His research is mainly focused on developing and effectively applying new detection methods based on techniques from robust statistics to detect potential money mules in an early stage. With only 0,002% of historically known mule accounts, there is a need to come up with innovative techniques to counter the ‘heavy class imbalance’ in the source data.
Tim Verdonck, professor of statistics at the KU Leuven and chairholder, speaks about the role that robust statistics can play in the detection of outliers in the data. The specific nature of transactional fraud requires that new methods be developed that are able to handle large data sets in which fraud is a very rare event.
Series of seminars
The stakeholders of the chair at BNP Paribas Fortis and KU Leuven wish for the success of the event to be the start of a Transactional Fraud Analytics community that could perhaps expand to other sectors. The chair offers a platform to organize a series of seminars where academic researchers and practitioners from different businesses can come together to share thoughts and ideas.