0

i am currently trying to replicate a predicitve churn model from SPSS modeller in R. When using SPSS, I used a 5.0 model and added a propensity score to each customer who was predicted to churn - I then used this to rank the highest propensity and used this to target the highest scoring customers first.

Now using RF in R, I;m starting to wonder if that was a vaid way to get a target list of customers. If I now have a list of customers in R who are predicted to churn - my question is who do I contact first as I cant contact them all? which of the customers are highest risk? How do I assign this score to each customer?

Many thanks

user3103335
  • 67
  • 6
  • 3
    Please provide a [minimal reproducible example](http://stackoverflow.com/a/5963610/496488). – eipi10 Oct 10 '16 at 21:46
  • 1
    The `randomForest` function returns the predicted probability of each potential outcome. If you have reason to believe that your model has sufficient predictive power, focus on the customers with the highest probability of churning. Or maybe you'd be better off focusing on the customers with a moderate probability churning--that is, perhaps your intervention has a better chance of preventing churn with those customers who are "on the fence" rather than with those who are set on leaving. Anyway, if you provide a reproducible example, we can provide more details. – eipi10 Oct 10 '16 at 21:51
  • Thanks eipi10. Haven't seen the random forest function spit out a probability for each prediction. Is it method = prob instead of class? will try and follow the above post on a reproducible example. Many thanks. – user3103335 Oct 10 '16 at 22:32
  • `predict(my_rf_model, type="prob")` – eipi10 Oct 10 '16 at 22:39
  • 1
    There's an R package that does C5.0 if that's a style of classification you prefer. – IRTFM Oct 10 '16 at 22:54

0 Answers0