Cluster Profiling Using Decision Trees#

Stage 7 · 🌳 Decision Trees · Lesson 48 of 56 · advanced

◀ Previous · How CART Decision Trees Model Interactions · Next · Using Decision Trees to Explain Clustering Results ▶ · ↑ Section

Important

✨ AI-generated content. This page was written with the assistance of an AI language model and is provided as a learning aid. Despite careful review, it may still contain mistakes, omissions, or out-of-date information. Whether you are new to the topic, a team lead, or a senior practitioner, treat it as a starting point rather than an authoritative reference: read it critically and independently verify anything you act on (code, commands, figures, and factual claims) against official documentation and primary sources before relying on it.

Clusters without descriptions#

Clustering (Stage 4) hands you groups, but not their meaning. k-means labels each customer with a cluster number, yet those numbers are opaque — cluster 3 is just “cluster 3”. Before you can act on segments, you need to describe them: what actually distinguishes cluster 3’s members from everyone else? This is cluster profiling, and decision trees are an elegant way to do it.

Turn labels into a target#

The trick is to turn the unsupervised result into a supervised problem. Take the cluster label each point received and treat it as the target to predict, using the original features as inputs. Then fit a decision tree to predict cluster membership. The clustering supplied the “answers”; the tree’s job is to find the rules that reproduce them.

Rules that define a cluster#

Because a tree is a chain of if-then splits (this stage’s opening lessons), the fitted tree reads as a description of the clusters. The path to a leaf dominated by cluster 3 might say: recency < 30 days and frequency > 10 — a plain-language profile of that segment. Each cluster gets a compact set of defining conditions, turning anonymous group numbers into interpretable characterisations a business can name and target — “recent frequent buyers”, say.

Why it works#

This works because trees bring exactly the right strengths: they are interpretable (the whole point here), they handle mixed feature types and interactions without fuss, and they naturally identify which features separate the groups — a built-in importance ranking. It is a recurring pattern in machine learning: use a transparent model to explain the output of an opaque one. The final lesson of this stage takes the idea one step further — using a tree as a surrogate to explain clustering results in general.

Hint