1 / 6
Formulate the problem
Decide what you want to learn and which variables to group on.
Group 12 café customers by visits per month and spend per visit to find customer types. Nobody comes with a correct type (label).
2 / 6
Distance measure
Decide how to measure "similar" as a number. The usual choice is the straight-line (Euclidean) distance.
3 / 6
Clustering procedure
Decide how to group. Hierarchical methods merge the two closest groups one at a time; nonhierarchical methods such as k-means fix the number of clusters first.
4 / 6
Number of clusters
Decide how many groups to keep. Stop just before the merge distance suddenly jumps.
5 / 6
Profile the clusters
Describe each cluster by its variable means (marked ×) and give it a name.
6 / 6
Assess validity
Check that the result can be trusted, by regrouping with another method or checking an index.