Hands-On Unsupervised Learning with Python

Ward's linkage

The last method we are going to discuss is called Ward's linkage (named after its author and originally proposed in Hierarchical Grouping to Optimize an Objective Function, Ward Jr J. H., Journal of the American Statistical Association. 58(301), 1963). It's based on the Euclidean distance and the formal definition is as follows:

At every level, all clusters are taken into account and two of them are selected with the goal of minimizing the sum of the squared distances. The process itself is not very different from average linkage and it's possible to prove that the merging process leads to a reduction in the variance of the ...