0
$\begingroup$

I am trying to develeop an algorithm with sklearn and Tensorflow to predict which car can be offer to each customer.

To do that I have a database with the answers of one survey to 1000 customers.

An example of questions/[Answers] are:

  1. Color/[Green,Red,Blue]
  2. NumberOfPax/[2,4,5,6,7]
  3. HorsePower/[Integer]
  4. InsuranceIncluded[yes/no/Don't know]

As you can see all questions are answer previously tipified, and in case the answer can be open I validate the value to be an integer or a radio button.

The purpose of that beahivour is that despite the categorical variables I can easily use sklearn to clustering the data.

Will be a good approach to translate this categories to numerical value as an intern procedure an then cluster with this references?

For example: yes --> 0; No --> 1; Don't know --> 2

Then sklearn will cluster with all variables as numerical values.

I have thought this possibility beacuse I believe that sklearn can not cluster nominal data.

What do you think about this approach?

$\endgroup$
  • $\begingroup$ Notice that the sample scale that you provide implicitly introduces properties, inherent to numbers that are not necessarily valid in your case. In particular you end up having total order and distance between categorical answers, which do not appear to make sense. Consider this explanation for more details. $\endgroup$ – mapto Aug 10 '18 at 12:40
2
$\begingroup$

Clustering on categories is not something sklearn can do by default. And assigning sequential values to categories like that certainly won't help - clustering tends to work based on distance, by assigning 0, 1, 2 to Yes, No, Don't Know like that, you are suggesting Yes is 'closer' to No than it is to Don't Know.

I highly recommend having a look at k-modes, a clustering algorithm for categorical data. Essentially it optimises according to how common the set of categories are within the cluster (modes, rather than the means of values). E.g. may find green people carriers (5/6/7 pax) in a cluster, and red sports cars (2/4 pax).

There is a Python library here, which also has links to some papers describing the algorithm. There is also k-prototypes, which clusters with combined numerical and catagorical data.

$\endgroup$
  • $\begingroup$ Thanks for your answer! I will check the algorithm. I guess that it can also cluster numerical data, right? $\endgroup$ – Maik Jul 20 '18 at 16:20
  • $\begingroup$ Yes, k-prototypes can cluster both by combining k-modes with k-means. $\endgroup$ – Ken Syme Jul 22 '18 at 8:28

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

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