Once we have built the classifier as above, and also know which clusters are likely to be important, we can search for events which belong to the clusters we are interested in. For example, from the rules shown in figure 5.6, we know that we only need to use cluster A to do the classification, thus there is no point looking for potential elements of clusters B or C, as they are not necessary for the classification made by the rule. This allows us to efficiently implement our event search. Of course, if time is not an issue, then you can default to extracting all possible events. However, by using this selective event searching, we can use our expectations to minimise the processing power required to do classifications for more complicated tasks, containing more than one channel and typically more than two classes.
The same issues that arose in event attribution occur here: namely, how do we assess whether a particular event has occurred? We can in fact use the event attribution process in place of the selected event searching process and the system will still function correctly.