Thursday March 27, 2:00 pm, room 05.001
Descriptive modeling and its applications
by Dragan Gamberger (Rudjer Boskovic Institute, Zagreb)
Descriptive modeling is application of machine learning for insightful data analysis that may result by elicitation of novel human knowledge. Subgroup discovery is a methodology based on rule induction that is appropriate for descriptive modeling tasks. The topic of the talk will be presentation of some results obtained on a recently finished EU project "FOC - Forecasting financial crises" and how the subgroup discovery methodology enabled us to come to these results. This case study will be used to illustrate important subtasks of descriptive modeling like supporting factor induction, handling imprecision of numerical attributes, and conversion of induced rules into risk models. The main results for economics are that we identified different subtypes of banking crises and demonstrated that one of these types is connected with problems with good governance, especially with corruption.
Dragan Gamberger is senior researcher at Rudjer Boskovic Institute, Zagreb, Croatia. His main scientific interests are inductive machine learning and knowledge representation for decision support tasks. The main achievements are rule learning based on the logic minimization approach, theory of feature relevancy, noise detection based on saturation, and rule learning for subgroup discovery. He is co-author of the book Foundations of Rule Learning, Springer, 2012.