Netezza Divisive Clustering

Divisive clustering is a method of cluster analysis in which the algorithm is run repeatedly to divide clusters into subclusters until a specified stopping point is reached.

Cluster formation begins with a single cluster containing all training instances (records). The first iteration of the algorithm divides the data set into two subclusters, with subsequent iterations dividing these into further subclusters. The stopping criteria are specified as a maximum number of iterations, a maximum number of levels to which the data set is divided, and a minimum required number of instances for further partitioning.

The resulting hierarchical clustering tree can be used to classify instances by propagating them down from the root cluster, as in the following example.

Figure 1. Example of a divisive clustering tree
Example of a divisive clustering tree

At each level, the best matching subcluster is chosen with respect to the distance of the instance from the subcluster centers.

When the instances are scored with an applied hierarchy level of -1 (the default), the scoring returns only a leaf cluster, as leaves are designated by a negative number. In the example, this would be one of clusters 4, 5, 6, 8, or 9. However, if the hierarchy level is set to 2, for example, scoring would return one of the clusters at the second level below the root cluster, namely 4, 5, 6, or 7.