Research Interests

My research focuses on the application of data analytics to real life business applications, such as in marketing and finance. A key element of my research entails the incorporation of profitability into data analytics, in order to align data-driven solutions with business requirements. This has led to the development of a profit-driven performance measure for customer churn prediction models and for credit scoring models. We are also working on the incorporation of profitability into the model building step itself.

My main research topics are:

  • Predictive analytics: classification models, (social) network-based classification
  • Profit-driven classification performance measurement
  • SOM-based clustering
  • Rule extraction
  • Bayesian network classifiers

I have worked on the following applications:

  • Customer churn prediction in the telecommunication and financial industry
  • Credit scoring
  • Software fault prediction
  • E-commerce acceptance

More information can be found on DataminingApps, or on the website of my research group, the Leuven Institute for Research on Information Systems (LIRIS).

PhD Project

Title: Business-Oriented Data Analytics - Theory and Case Studies

Doctoral committee:

  • Prof. dr. Bart Baesens (KU Leuven)
  • Prof. dr. Marnik Dekimpe (Tilburg University)
  • Prof. dr. Theodoros Evgeniou (INSEAD)
  • Prof. dr. ir. David Martens (Universiteit Antwerpen)
  • Prof. dr. Martina Vandebroek (KU Leuven)
  • Dr. Bram Vanschoenwinkel (AE)

The public defense found place on September 12, 2013, in Leuven, Belgium.

This PhD thesis focuses on predictive analytics in a business environment. Unlike explanatory modeling, which aims at gaining insight into structural dependencies between variables of interest, the objective of predictive analytics is to construct data-driven models that produce operationally accurate forecasts.

Such a predictive analytics tool consists of two components, (1) data-driven models designed to predict future observations and (2) methods to assess the predictive power of such models. This dissertation focuses on a sub domain of predictive analytics: binary classification. Hence, two components are of interest: the classification models themselves, and the classification performance measures. We argue that profitability should be integrated into both components.

We propose an approach which looks at benefits and costs, instead of misclassification costs alone. By focusing on benefits (and profit) rather than costs, we are staying closer to the business reality, and aid the adoption of classification techniques in the industry. Therefore, a profit-based classification performance measure is developed and applied to real life business cases. Moreover, an exploratory study on the incorporation of the profitability criterion into the model building step is presented.

Finally, this PhD thesis discusses two case studies which clearly demonstrate the usefulness of data analytics in a business context.