Me when I get an email from our Chief Data Scientist that starts off like this….
Reklaim uses an Ensemble Model to predict a client’s customer account risk and to set appropriate account credit limits. Our current model was developed after months of testing various Ensemble technologies including but not limited to bagging, boosting, stacking, voting, and random trees.
Our Prediction Method is called an Ensemble Model because the model development process creates a number of neural networks and then weights and combines them to produce a “strong” model (the Ensemble) that is more predictive than any one of the individual models it is created from.
The Prediction Method is run numerous times using different random seeds to determine the various Test and Validation samples. The model that produces the smallest Validation Prediction Error (its PMML file) is used as the basis for determining each account’s Credit Score, its Credit Limit, and predicting which accounts will remain GOOD and which accounts will become BAD in the next period.
And yes, believe it or not, this is the simplified version!
The one written for those of us who hear “random seeds” and think about a “bag”‘from the Dead lot in the mid 80’s.
What you’re reading has actually been scrubbed of several specific variables to prevent reverse-engineering our secret sauce, which means there’s a denser, scarier version of this email floating around out there. I have decided, for the sake of my own sanity, that I will never open it.
To our Data Science team: I trust you completely. Please keep doing whatever it is you just said.