The value of customer behavior data in predictive modeling

Date: 15 April 2016

Venue: University of Antwerp, Hof van Liere - F. de Tassiszaal, Prinsstraat 13 - 2000 Antwerp

Time: 5:00 PM

PhD candidate: Julie Moeyersoms

Principal investigator: Prof David Martens

Short description: PhD defence Julie Moeyersoms - Faculty of Applied Economics


In this world of ubiquitous computing and increasing connectivity, people's digital footprints continuously grow. This creates a new, fine-grained type of data that reveals more detailed and low-level human behavior (such as payments, website visits and phone calls). This data is different from other, more traditional data types (such as socio-demographics) in a way that it captures people's actions, opinions, thoughts and social interactions. The aim of this dissertation is to investigate the value of such fine-grained behavioral data in predictive modeling.  In the first part, value is approached in terms of predictive performance. We research whether adding this type of data improves our predictions on consumer behavior. Several real life data sets are researched, going from marketing applications such as churn prediction or online advertising, to risk management applications such as residence fraud detection and credit risk predictions. 

The second part of this dissertation approaches the economic value of fine-grained behavioral data. More specifically, we look at the application of online advertising where browsing behavior data is used in order to infer a user's interest in a certain advertisement. Typically, hundreds of thousands of websites are included in the model, each adding value to the predictive performance of the model. In this work, different methods are proposed to distribute economic value to each of these data sources to compensate for their value in the prediction model. Ultimately, this allows for better insights in the actual added value of a data source, which can be valuable information for practitioners when deciding on data acquisition prices or when evaluating their data-driven decisions. Without the tools to make such evaluation, big data is arguably more a faith-based initiative than a scientific practice.