3
$\begingroup$

I have historical data from an e-shop transactions. I want to write a prediction model and check if a specific user will buy with or without a discount, so I can do some targeting offers.

The idea is:

  1. If a user will buy the regular price, will not have an offer.
  2. If a user will not buy the regular price, check if he/she will buy with an offer.

With this way, I will avoid to make an offer to someone who would buy with the regular price.

So, I am still in the brainstorming and trying to find a way for implementing the 1-2. Should I create two separate models to predict the 1) and then the 2) with the second model? Or should I join both in one prediction model?

$\endgroup$
  • $\begingroup$ Did any of the answer below help you? $\endgroup$ – Dawny33 Nov 20 '15 at 16:52
0
$\begingroup$

You can use Decision trees for a single model prediction for both the set of users.

A good start would be to first read up on Decision Trees and their applications.

You can include the offer as a decision(as a boolean in this case).

buying with offer and buying without offer can be the decision criterion.

You can, in fact go ahead and put in the offer values also. For example,

offer>10% and offer <10%

$\endgroup$
  • $\begingroup$ Sounds reasonable since I have the same data for both cases. However I am not sure how it will work for users that haven't seen an offer before. Neither accept or reject one. I was hoping to do it through similar users with about the same features. Is it possible with decision trees? $\endgroup$ – Tasos Oct 12 '15 at 10:20
  • $\begingroup$ Yes, it is possible. Please look up the read up hyperlink in the answer. $\endgroup$ – Dawny33 Oct 12 '15 at 10:24
  • $\begingroup$ Sorry, I was on my phone and didn't see that there was a link on your answer. I will check it and if it actually makes sense, I will choose it as the best answer :)...Thank you $\endgroup$ – Tasos Oct 12 '15 at 13:15
0
$\begingroup$

Your requirements suggest you might want to use uplift Modelling. This has a neat R package references in the wiki article.

This looks like a practical paper.

$\endgroup$

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.