Consider you are building a score model, (ex: given an amount of personal information about your clients, you want to know what is the approximate effect of each variable at predicting. For example, the probability that the client is not going to fully pay for their purchase) .
Now consider that, after having this model, you also want to add some 'rules' in the sense that if the client fails to comply with this rule you automatically predict that he will not pay for its purchase, hence this would immediately make the predicted variable to be 'client is not going to pay for their purchase'.
Does anybody has a solution to put this entire 'model' running in DataRobot?
I thought about running sequential models, one for each rule where I would feed them with fabricated data to force the perfect correlation between the rule and the outcome and another one for the score.
Solved! Go to Solution.
Generally, when people use business rules on top of their modeling scores, this is done outside of DataRobot. For the customer payment example, you can have the modeling scores added to a database as a probability between 0 and 1 that the customer will not pay, along with a "will pay"/"will not pay" classification feature.
Within that database or table, you could then track the rules that you want to add to the modeling results. Something like, the customer did not respond to the prior 3 emails. You could then write some simple logic: customers who have a probability to not pay > 70%, AND they didn't respond to prior 3 emails THEN overwrite the classification to "will not pay".
Does this make sense?
Thank you Emily.
Inversely, I was trying to understand if there was any possibility to have everything inside DataRobot, so that the entire process would be done until the deployment phase.
However, I do understand the best method would be definitely the one you explained.