Exceptional Model Mining
Date: 2011-01-27 Time: 11:00 Room: [[http://www.cs.uu.nl/docs/reach/bbl.php][BBL]] [[http://www.cs.uu.nl/info/plan/bbl.php]] ----+++++ Speaker: Matthijs van Leeuwen ----+++++ Title: Exceptional Model Mining ----+++++ Abstract Companies nowadays have plenty of data, collected everywhere and stored in data warehouses. How to extract valuable knowledge from this data is usually not straightforward; that is where the field of data mining comes in. Data analysts are often interested in subsets of the data that are `interesting' with regard to some specific property. This is exactly what can be mined using Subgroup Discovery methods; a subgroup is a subset of the data for which the data distribution of the target property deviates substantially from that of the subgroup's complement. In this talk, we will first discuss Subgroup Discovery and then explore its generalisation, Exceptional Model Mining. This generalisation allows for multiple target properties instead of just one target property. The results of the experiments I will show clearly reveal that this leads to new possibilities for the analysis of complex data. Although good results have been obtained so far, interesting algorithmic challenges remain.