July 7, 2019
Around 150 teams of students from 28 countries took part in the international “DATA MINING CUP 2019”. Alexander Melde, Lukas Theurer und Christian Wernet, three students undertaking a Master’s degree in Computer Scienceat Karlsruhe University of Applied Sciences, were amongst them. Their entry was awarded an excellent sixth place. They also held their own against teams from renowned German universities, coming third amongst them.
As one of the ten best student teams they were invited to the “Retail Intelligence Summit” in Berlin on July 3 to present their work. They were also present at the prudsys AG award ceremony and the celebrations afterwards.
“Data Mining” is the systematic process of using statistical methods to identify new trends or cross-references within large databases (also called mass data or “big data”).
When competing in this year’s edition of the international students’ competition, which also marks its 20th anniversary, the participants were tasked with uncovering instances of fraud occurring when self-scanning terminals in supermarkets are used. The trend towards the use of such self-scanning devices in supermarkets, which allow customers to scan their purchases using a smartphone and the supermarket’s app, for example, and to place them directly in their shopping carts and pay for them, is growing as it can reduce the amount of time customers have to wait in line at the checkout. However, in 5% of cases, investigations have revealed inconsistencies. Whether these are caused by deliberate acts of theft or fraud, or whether faults within the smartphone app are to blame, is as yet unknown.
The students were tasked with developing an algorithm that uses anonymized purchase information to recognize whether or not fraud is being committed. The term ‘fraud’ is used to mean all forms of inconsistency – both deliberate theft and system errors.
The team from Karlsruhe University of Applied Sciences used machine learning methods to develop a means of processing data, such as time spent shopping, number of products scanned and their value. As a result, they were able to identify inconsistencies based on inconspicuous properties of a purchase. A neural network was also used for further data processing, which then “judges” whether or not fraud has been committed by the customer. It wasn’t only important for them to develop an especially effective system, they also wanted to avoid labeling innocent shoppers as thieves or fraudsters. The system also has to be particularly reliable – an important quality as the ethics of using artificial intelligence (AI) is currently a matter of public debate.
The team became aware of the DATA MINING CUP after attending a lecture on “machine learning” held by Astrid Laubenheimer, Professor of Engineering at the Faculty of Computer Science & Business Information Systems. Both the Faculty and the university’s Intelligent Systems Research Group (ISRG) supported the team, not only with IT infrastructure, but also with technical discussions.
To read more about the DATA MINING CUP, visit www.data-mining-cup.com