Practical Predictive Analytics

Hierarchical clustering

As an alternative to k-means, you can perform hierarchical clustering, which does not require you to specify the number of clusters beforehand. In hierarchical clustering, you can either start with one cluster and then continue to subdivide two clusters at a time, or start with every record being in its own cluster, and then merge them together to become a larger cluster. In any case, it is easy to scan up and down the tree, known as a dendrogram, and identify the grouping of clusters that suit your needs, rather than having to run the clustering one at a time. However, for large computational processing, partitioning clustering (k-means) definitely has a performance edge, which is why it is favored in industry.