Clustering and Segmentation

What is clustering?

Clustering groups observations by similarity in their features, without a known target label (unsupervised learning).

RFM (Recency, Frequency, Monetary value) is a simple, popular framework for customer segmentation and can be used with or without clustering.

Example

A retailer clusters customers using purchase frequency, average spend, and category mix:

These labels must be interpreted from the actual results, not assumed beforehand.

How it works

k-means steps:

1. Choose the number of clusters, k
2. Place k starting centres
3. Assign each record to its nearest centre
4. Move each centre to the average of its assigned records
5. Repeat steps 3–4 until assignments stabilise

Other methods: hierarchical clustering, DBSCAN (finds irregular shapes and flags noise), Gaussian mixture models.

Evaluate usefulness

A mathematically neat segmentation isn't valuable if the groups can't be explained or acted on.

Common mistakes

• Treating clusters as natural or permanent "types" of people.
• Forgetting to scale features.
• Choosing k purely by a statistic without considering business use.
• Using too many features, which makes clusters hard to interpret.
• Never re-running: customer behaviour changes, so segments need refreshing.

Try it: Choose three numeric features for customer segmentation and explain why each one may help distinguish different usage patterns.