2
$\begingroup$

I am attempting to build a model that predicts the likelihood of 1000 customers churning every week, for the next 5 week. My training consists of 4 continuous feature variables, and a class variable that represents whether or not a customer churned in the upcoming week, which Churn being defined as cancellation of service.

'data.frame':   1000 obs. of  5 variables:
$ ID             : chr  "9722209" "9722213" "9722215" "9722223" ...
$ feat.1         : num  2 5 1 2 7 2 0 5 2 2 ...
$ feat.2         : num  3 2 2 3 1 2 2 6 4 9 ...
$ feat.3         : num  9 4 1 2 2 8 2 2 2 2 ...
$ feat.4         : num  2 0 0 0 2 5 4 2 0 0 ...
$ churn.7.days  : num  1 0 1 0 0 0 0 0 0 0 ...

My question is about this: how can I use this data set to predict the likelihood of churn for not only the next week (which is relatively straight forward), but for the next subsequent 8 weeks?

$\endgroup$
  • $\begingroup$ Welcome to CV! Can you define "churn" for us? You can edit your question to make that clear by clicking the "edit" link in the lower left. $\endgroup$ – Alexis Apr 12 '18 at 20:49
  • 1
    $\begingroup$ Thank you, it's good to be here! I have edited the definition of churn above. $\endgroup$ – ari8888 Apr 12 '18 at 20:55
  • $\begingroup$ Is time a variable in this data set? $\endgroup$ – Alexis Apr 12 '18 at 22:11
  • $\begingroup$ It isn't. The only time information we have is within our class variable, which is the event of cancellation within the upcoming week... $\endgroup$ – ari8888 Apr 12 '18 at 23:02
1
$\begingroup$

Because your outcomes are one week churn events I do not think you data support analysis of 8 week conditional churn probabilities without some assumptions having a serious likelihood of being false. (I.e. successive churn probabilities are (1) independent, and (2) homogeneous.)

tl;dr: You can't "predict the likelihood of churn for not only the next week (which is relatively straight forward), but for the next subsequent 8 weeks"

$\endgroup$
  • $\begingroup$ That makes sense to me. Out of curiosity, if we were to break the assumptions of independence and homogeneity, could we use the probability of churn within the next week as a prior to a calculate a posterior probability model that predicts future churn? $\endgroup$ – ari8888 Apr 13 '18 at 15:10

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.