0
$\begingroup$

I have a dataset related to real estate Market with features as price, area, number of bedrooms, bathrooms, and zip code. I want to group them in 4 different groups for each zip code. So, I used k-means clustering with 4 groups. Now, how can I explain the results to the business team? They are asking:

  1. How reliable is this grouping?
  2. How did I decide the threshold while grouping them together?

It makes me wonder if there exists any other way by which they can be grouped together.

Improve this question
$\endgroup$
5
  • 2
    $\begingroup$ The business team are asking good questions! Many statistical people here would like to swap their business team with your business team. $\endgroup$
    – Nick Cox
    Commented Apr 16, 2019 at 10:02
  • $\begingroup$ What kind of grouping do you want to have? What information do you want out of it? What kind of business decision is it supposed to support? There are an infinite ways of grouping this, the approach depends on what you want to achieve... $\endgroup$
    – Jon Nordby
    Commented Apr 16, 2019 at 13:18
  • $\begingroup$ @jonnor I want to group the houses into 4 different groups for targeting different groups of customers so that we can send promotional offers to them. For example group 4 can be a rich customer ( he is more interested in Expensive and big houses) as compared to group 1 consists of relatively poor people ( Less expensive and small houses ). One way of doing is to Hardcore define threshold with help of business ,such as <100k , 2 bed, 1 bath as group 1 similarly for others. I was wondering if it is possible with Clustering without hardcore defining them. $\endgroup$
    – No_Body
    Commented Apr 16, 2019 at 13:41
  • $\begingroup$ What you are talking about is customer segmentation/targeted marketing/product suggetions. If you want to learn that from data, you should probably learn it from customer preferences (demand side) - not from housing data (supply side). Typical data used would be typical searches etc. Once you know what potential customer (groups) are interested in, then you can just filter the housing data based on that. $\endgroup$
    – Jon Nordby
    Commented Apr 16, 2019 at 15:01
  • $\begingroup$ Unfortunately, I don't have the data from demand side. That's why I was forced to work with housing data. $\endgroup$
    – No_Body
    Commented Apr 18, 2019 at 12:41

1 Answer 1

Reset to default
3
$\begingroup$

It's likely to be complete statistical nonsense.

Because all variables are different. And computing differences of zip codes is pretty much nonsense.

Obviously, a difference of 1 in the zip code is not equivalent to a difference of 1 in the number of bedrooms.

It's not enough to just throw an algorithm at some data, but the data needs to be prepared in a way that the right problem is optimized. So k-means did its job - but you asked the wrong question. "How do I minimize variance in zip codes plus bedrooms?" is not a good question.

Improve this answer
$\endgroup$
5
  • 1
    $\begingroup$ Hi, You're right about the zipcodes. I only used Price, area, no of bedroom, bathrooms after normalization for clustering. I computed 4 clusters for every Zip code.I didn't include zip codes in the data itself. $\endgroup$
    – No_Body
    Commented Apr 12, 2019 at 14:34
  • 2
    $\begingroup$ Nevertheless your features have very different meanings. So what is the meaning of the function you optimized? $\endgroup$
    – Has QUIT--Anony-Mousse
    Commented Apr 16, 2019 at 9:36
  • 1
    $\begingroup$ Even normalization (whatever that means precisely) isn't guaranteed to make k-means defensible. $\endgroup$
    – Nick Cox
    Commented Apr 16, 2019 at 10:00
  • $\begingroup$ Thanks for the replies guys. I highly appreciate it. @Anony-Mousse, apologies, but when you say very different meanings. Does this mean they should have same measurements. For example, If i have similar features like (no. of bed, no. of bath, no. of garage) OR (area of bed, area of bath, area of garage) then I can cluster them ? The data-set given in question has mix of all kinds of features so i can't use K-Means here?Sorry I am still kind of new to this area. $\endgroup$
    – No_Body
    Commented Apr 16, 2019 at 14:00
  • 1
    $\begingroup$ You can use whatever you want. And you can interpret what you can interpret. That sounds banal, but the point is that this is a choice that you make, essentially how to weight one bed squared vs. one pool squared vs. one dollar squared. There is no right. But you may or may not be able to argue that the result is meaningful for your use case. $\endgroup$
    – Has QUIT--Anony-Mousse
    Commented Apr 16, 2019 at 18:26

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

Not the answer you're looking for? Browse other questions tagged or ask your own question.