[PaulHoule]

It reminds me of the old PRISM system described here:

https://www.amazon.com/Clustering-America-Michael-J-Weiss/dp...

which was not an unsupervised clustering but rather a grid subdivision into communities over a few variables. Then they gave catchy names like "Shotguns and Pickups" and "Blue Blood Estates" to the boxes.

The current study divided first into three categories of growing, stable and shrinking and then split the growing communities into high, middle and low income.

That kind of division is more likely to be meaningful than an unsupervised clustering (e.g. I can explain the structure in a sentence so of course it is meaningful.)

[nickt]

There are a few of these systems, my first exposure (I was based in the UK, so these systems lean to UK geodemographic types) and use of them was back in the early 1990’s with what was called CCN Mosaic at the time. Mosaic is now owned by Experian [1]. Another similar system, ACORN [2], was developed by the same guy, Richard Webber [3].

I was running reasonably complex models of household types around stores to build marketing campaigns for consumer goods. It was very effective.

I met Prof. Webber a couple of times, and he explained that the genesis for these systems was for siting public transport stations in the right places to optimise usage patterns and ensuring they had sufficient usage to make the investment worthwhile, well, IIRC, that was 30 years ago…

[1] https://en.wikipedia.org/wiki/Mosaic_(geodemography) [2] https://en.wikipedia.org/wiki/Acorn_(demographics) [3] https://en.wikipedia.org/wiki/Richard_Webber_(demographer)

[PaulHoule]

I had fun years ago working for a company that developed models (usually of a different kind) for deciding where to locate retail stores in space or how to allocate sales territories in space.