Check for overfitting
A very high in-sample AUC like \(99.9\%\) can be an indicator of overfitting. It is also possible that your dataset is just very well structured, or your model might just be terrific!
To check which of these is true, you need to produce out-of-sample estimates of your AUC, and because you don't want to touch your test set yet, you can produce these using cross-validation on your training set.
Your training data, customers_train
, and the bagged tree specification, spec_bagged
, are still available in your workspace.
This exercise is part of the course
Machine Learning with Tree-Based Models in R
Exercise instructions
- Using
fit_resamples()
, estimate yourroc_auc
metric using three CV folds of your training set and the model formulastill_customer ~ total_trans_amt + customer_age + education_level
. - Collect the metrics of the result to display the AUC.
Hands-on interactive exercise
Have a go at this exercise by completing this sample code.
set.seed(55)
# Estimate AUC using cross-validation
cv_results <- fit_resamples(spec_bagged,
___,
resamples = vfold_cv(___),
metrics = ___)
# Collect metrics
___(cv_results)