If testing reveals that old classes are constantly being overwritten due to the creation of new dense clusters, which are assigned a new class id if "Unclassed" per algorithm 2 (Paper 3, page 10), then we might need to change the classification step to only occur during the final cluster formation - i.e. assign the class of the dense micro-cluster closest to the final cluster centroid to all micro-clusters in that final cluster, perhaps only creating a new id if that micro-cluster has no class (credit to @piratevector for this idea).
If testing reveals that old classes are constantly being overwritten due to the creation of new dense clusters, which are assigned a new class id if "Unclassed" per algorithm 2 (Paper 3, page 10), then we might need to change the classification step to only occur during the final cluster formation - i.e. assign the class of the dense micro-cluster closest to the final cluster centroid to all micro-clusters in that final cluster, perhaps only creating a new id if that micro-cluster has no class (credit to @piratevector for this idea).