Define scenarios on predicted output for prescription

zain

I am working on telecom churn use case in rapidminer and have built the model using logistic regression. After executign the model I am getting the prediction successfully.Now I want to recommend some actions or prescriptions to stop customer churn.for example I want to output such text based on the predicted data:"increase internet volume for old users in same price","improve fiber optic service".In order to achieve such output I have selected this approach as follows:

1) define business rules on the predicted outcome data (like if customer minutes are less than <140 and internet usage is 5GB then recommend bla bla package).

2)by covering all the use cases based on the predicted data show different outcomes in text to the client to stop customer churn.

The way to do this is I guess through coding in a language that is supported by rapidminer.

I want to know how can I achieve this functionallity. If you have any questions please feel free to ask.

Thank you.

MartinLiebig

Hi,

i think there is no need for additional software.
You would model the likelihood of churn for given parameters (changeable and unchangeable). Then you vary the changeable parameters and calculate the likelihood of churn again (e.g. via Optimize Parameters). Afterwards you get the action which reduces the likelihood of churn the most. You can report this back to business.

BR,
Martin

Telcontar120

You can also use operators like Tree to Rules to convert decision paths into rules that can be communicated in text.  You can even do this with outputs from another learner by creating a categorical version of your prediction if you want.