I have a data set containing many client's id, and its behavior characteristics measured each month before churn or censored. Data looks like:

id || lifetime period || folow-up time before churn of censores || churn or censored || large list of behavior variables ||

Each id exist for many times in my dataset during its lifetime.

I want to build model that allow me to calculate churn probability for each client for each its future lifetime period and update this probability each month during clients lifetime. I started with Cox regression with time-varying covariates but further realized that it is inappropriate for predicting purposes. Now I started building Cox proportional hazard model.

It is appropriate choice for my purposes? Should I build new model for each lifetime period? Or I can include in train sample many observation for each ID but with different covariates and different folow-up time?


Cox Proportional Hazard model is not really appropriate for churn modeling because the hazards are almost never proportional to each other in practice (they cross each other over time.)

You should consider using Discrete-time Flexible Hazard model using logistic regression. Check out the summary of this method in the following paper: http://www.lexjansen.com/nesug/nesug05/pos/pos6.pdf

There are more references mentioned in the link above, which should be helpful for further background information. SAS used to offer a course on this topic (with instructor Will Potts) -- not sure if it's still offered. You can check it out if you're interested.


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.