Question: You teach both university students and professionals. What analytical concepts do you think are the most important for analysts to have a solid understanding of?
Bart: First and foremost, I think an analyst should be aware of the need to have high-quality data before any analytical project can be started.
Next, he/she should have a solid understanding of both predictive (e.g., regression, classification) as well as descriptive (e.g., clustering, association/sequence rules) analytics.
The analyst should know the techniques underlying, how to appropriately use and evaluate them, and also understand how the resulting quantitative models can be successfully embedded into a business environment (e.g., marketing, finance, fraud, risk). Finally, the development of a model monitoring and backtesting framework should also be properly taken into account.
Question: You've been an advocate for applying survival analysis methods to customer events of interest. Do you see adoption of these methods increasing across industries?
Bart: In fact, I do. Many firms nowadays use classification models to predict: e.g., loan default, churn or fraud. These models have typically been built using either logistic regression or decision trees. The obvious next step is to extend these models to not only predict if an event (e.g., default, churn, fraud) will take place, but also when it will take place. Survival analysis provides us with an excellent tool suite to do this since it allows us to predict the timing of events.
Question: How are you and your fellow staffers incorporating more visualization in your analysis and/or in the presentation of results?
Bart: We extensively use visualization during the preprocessing step of an analytical project since it allows us to have a basic feeling about potentially interesting patterns present in the underlying data. We also use visualization more and more in postprocessing as well since analytical models should not only be high-performing but also understandable. Appropriate visualization facilities are key to represent analytical models in a user-friendly way.
Question: What new projects are you working on?
Bart: We are working on various projects at the moment. From a technical perspective, we are currently working on evaluating analytical models in a profit-driven way, rather than a statistical way based on lift charts, ROC analysis, etc.
We are also developing new social network techniques to do both churn and fraud detection. From an application perspective, we continue our work on credit risk and marketing analytics, but also explore new application areas such as business process analytics, whereby starting from an event log of an information system the underlying process model is being discovered.
Join us for the webcast July 17
Bart Baesens is a consultant and an assistant professor in the Research Center for Management Informatics at University of Leuven. During this webcast premiere, he will be discussing some of his extensive research on predictive analytics, data mining and customer relationship management — including the latest on survival analysis, scaling analytic efforts and more.
What else should you know about Bart? My favorite facts are that has a full-size UK phone booth in his garden in Belgium and that his favorite American food is key lime pie.