I'm doing some data prep on a dataset provided by a telecommunication company. There is a continuous variable that indicates how many months have passed since a customer renewed her contract. However, 20% of observations have missing values. I figured out that there are two reasons why a value might be missing:

1- a customer did not renew the contract

2- a customer is still on her first contract and therefore never had the chance to renew it.

What I want to do is to fill the missing values with respectively "Not renewed" and "Not applicable", and then discretize the rest of the numerical values through a supervised algorithm (chi2, MDLP) so that the variable would turn into a categorical one.

In the end, for this variable I'd have some observations categorized through simple labeling, while others through a supervised discretization algorithm.

Question: is it correct to do so? If not, how should I handle the problem?


Please see my comment below for additional information.

  • $\begingroup$ could you share some info on what sort of analysis you want to carry out in the end? That might give a better idea of what to suggest. $\endgroup$ – Marcus D Apr 27 '16 at 16:54
  • $\begingroup$ Ditto @MarcusD comment... Why are you doing this? Are you trying to make it more human readable for management, are you feeding this into a learning algorithm, or are you trying to produce insightful intelligence using statistical inference? Our answers will depend upon the purpose. $\endgroup$ – AN6U5 Apr 27 '16 at 17:39
  • $\begingroup$ I'm going to use this variable into a classification algorithm (probably decision tree) where the target variable is a binary 0/1 that identifies whether or not a customer contacted the call center in a given month. I want to discriminate between those who did not renew the contract and those who are still on their first contract and never had the chance to do so. This information is important because I believe that those who are not renewing might indeed be unsatisfied customers who therefore are more likely to contact the call center. $\endgroup$ – Davide Apr 28 '16 at 9:06

Your Answer

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

Browse other questions tagged or ask your own question.