Prediction of person characteristics with behavioral data
The University of Antwerp has developed advanced data mining and big data techniques to automatically predict persons’ interests and intents. This technology enables companies to find prospects, limit churners and avoid fraud.
Data mining is the process of automatically finding patterns in data sets. It has a long history. It has been used by banks to find patterns that describe how their good customers differ from the defaulters, by governments to find patterns that describe fraud, and by marketeers to find the most valuable customers. Recently, behavioral data has become available: data on actions taken by persons. Such data consists of little bread crumbs that persons leave behind in the digital world and is very predictive for persons’ interests. Think of Facebook pages liked, webpages visited, accounts paid to, contacts called with, etc. Finding patterns in such data requires tailored algorithms, which are being developed at the Applied Data Mining group.
Data mining researchers from the University of Antwerp have developed algorithms that use massive behavioral data to make predictions. These algorithms have been used to help banks leverage their already existing data assets (payment data) to better assess the credit risk of loan applicants. Similarly, in collaboration with a micro finance lender, Facebook like data have been used to predict credit worthiness,with great accuracy. UAntwerp spin-off Predicube uses the webpages that one looks at to predict which products persons are most likely interested in, which are subsequently shown as an ad for that product. In all cases, the value of behavioral data is tremendous, leading to improved marketing and risk management practices. Interestingly, the more behavioral data is available, the more accurate the predictions become. This means that large publisher groups, with more users and more webpages are better equipped to predict which ads are most suited for a person. Also, bigger banks, with more customers and more payment data, can better predict credit and fraud risk. This “bigger is better” approach should motivate large companies to start using their behavioral data, and smaller companies to consider pooling their data. In this research, the principle of privacy by design is used, trying to include privacy concerns in the algorithm. Also, the ability to explain such models to decision makers is of crucial importance as well. On the latter methodology, the University of Antwerp has obtained a patent, together with New York University.
About the researchers
The Applied Data Mining research group was founded in 2011 by prof. David Martens at the Faculty of Applied Economics. Their research focuses on two tracks: mining behavioral data and explanation algorithms for prediction models. This work has been published in highimpact journals, and was awarded several prices, including the “European ResearchPaper of the Year” by CIONET in 2017. The group works closely together with the private and public sector and was at the origin of the Predicube spin-off, active in ad tech.
De UAntwerpen maakt gebruik van cookies en daarmee vergelijkbare technieken voor het functioneren, meten en optimaliseren van de website. Ook worden er cookies geplaatst om b.v. YouTubefilmpjes te kunnen tonen en voor marketingdoeleinden. Deze laatste categorie betreffen de tracking cookies. Uw internetgedrag kan worden gevolgd door middel van deze tracking cookies Door op 'Accepteer alle cookies' te klikken gaat u hiermee akkoord. Lees ook het UAntwerpen Privacy statement