The event attribution stage uses the synthetic events created from the parameter clustering and then looks at each training stream to see if it has an event that is similar to the cluster we are considering.
Consider the Blues and Reds domain. We can now ask whether any of the training streams have events corresponding to clusters A, B and C as shown in figure 5.5. The results are shown in table 5.3.
It is possible to be more general and clever. Depending on the clustering algorithm, we may retain information from the parameter clustering that would provide information about the types of variations we expect. For example, we might be able to deduce that a delta event of (6,2) is still, most probably, a member of the cluster A. Thus, rather than using a simple true/false result, we might instead be able to estimate the confidence that a given stream has an event belonging to cluster A.
Thus, we can treat each of the clusters as a synthetic event, and we can evaluate the confidence for each stream that it has an event corresponding to that attribute.