In our experiments we have decided to use M5 [14, 15]. M5 is a propositional learning system that is able to deal with numeric classes and learns model trees [14]. Model trees differ from regression trees [2] in that they have linear models in their leaves instead of average values. The algorithm for building model trees is similar to the one used in building decision trees. However, M5 is a kind of hybrid algorithm in the sense that it combines instance-based learning with model trees. It can learn an instance-based model of the data, a model tree or a combination of both. In its default mode the system decides which methodology to use for a given problem. In our experiments we used all default parameter values.