Market / Customer Segmentation - Merging two different segmentations

Question

I have a database where each observation is a person. They were questioned on their attitude towards the consumption of X category of product. I have being using K-means to segment this data.

I have noticed that people under 19 years old tend to be quite different in their responses to those over 19. I was thinking of dividing the data in <19 and >=19 and producing two clusterings and then merging them so as to produce a single report.

Does this make any sense from a data mining point-of-view? Is there precedent for doing so?

Answer 1

If you are using the clustering algorithm correctly, it should already discover something like this. Other than that, it sounds more like decision trees to me than clustering! Maybe you'll want to build a decision tree with early pruning, and then use the leaves as "clusters"?

As your data probably is discrete, k-means will have a systematic error. K-means is best for truly continuous vector data, where each dimension should have the same value range and semantic scale.

Definitely pay attention to your cluster sizes. If you are seeing empty or near-empty clusters, your data is not working for k-means!