How to build a confidence interval in XGBoost?

HowTo
#1

Hi All,

Can anyone offer me advice on how to build a confidence interval for the results of an xgboost (binary) model?

I’ve researched this question on the web, but the answers require R skills beyond my level. Is there anything built into Exploratory (awesome software btw!) that would let me calculate CIs natively?

Thanks.

#2

Hi,
Unlike logistic regression or linear regression, confidence interval is not calculated as a by-product of prediction in tree based algorithms like XGBoost or RandomForest.

One way to indirectly estimate confidence interval I found is by experiment, by changing random seed and repeatedly predicting, though it would still require writing some R…