This is not easy to determine a priori. At a minimum, meaning can be extracted in the following ways: firstly, the features that are selected by the feature selector indicate which event clusters and global features are useful, hence this tells us something about which parts of a stream are important for classification purposes. Secondly, if the learner produces intelligible output, such as a rule builder, we can find out how the various events are related and this build an understandable model by combining the rules with the descriptions of the synthetic events. For example, it would be possible to convert from a rule description and the cluster information back into a prototype or set of prototypes for each class.