군집 프로파일링과 이름 붙이기

An algorithm hands you labels like 0, 1, and 2. Those numbers mean nothing until you profile each group and explain it. 🏷️

Start by adding the predicted labels back to your DataFrame as a new column, so every row knows its cluster.

Now group by that column and average the features. Each clusters profile appears as a row of typical values.

Scan those averages for what stands out. The features where clusters diverge most are the story you will tell.

Count members per cluster too. A tiny cluster may be a niche segment, or a sign your k was set too high.

A quick scatter plot colored by cluster shows whether the groups are distinct or blur into one another.

Translate each profile into a human label, like high spenders or weekend browsers, so stakeholders grasp it instantly.

For every segment, write one plain sentence describing who they are and what makes them different from the rest.

If a group has no clear traits, it may be noise or overlap. Consider a different k or algorithm before naming it.

The point of naming is action. A segment called price-sensitive can guide a discount, while loyalists earn a reward.

Clusters can drift as data changes, so re-profile periodically to make sure the names still fit the people inside.

Let us check the real goal of profiling clusters.

Profile clusters by comparing their feature averages, then name each segment so the groups drive real decisions. 🎯