I'm modeling how likely a song is to occur in a playlist with a particular title. I can calculate a simple probability based on my current data, but it's highly zero-inflated because my data is limited. I've been recommended a few ways to tackle this, but am struggling to understand which one is optimal and why. So far:

  • Hurdle (two stage) model
  • Tweedie
  • Zero-inflated Gamma
  • Some type of log(1+y) transformation

Any thoughts on how I should approach this? My data is also typically about 95% zeros for any given title, and most of the remaining 5% is a low positive decimal lower than 1.

  • $\begingroup$ Can you post some sample data? $\endgroup$ – user2974951 Sep 10 '19 at 8:55

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.