Obtain variable importance scores

Usage

# S3 method for bagger
var_imp(object, ...)

Arguments

object: An object.
...: Not currently used.

Value

A tibble with columns for term (the predictor), value (the mean importance score), std.error (the standard error), and used (the occurrences of the predictors).

Details

baguette can compute different variable importance scores for each model in the ensemble. The var_imp() function returns the average importance score for each model. Additionally, the function returns the number of times that each predictor is included in the final prediction equation.

Specific methods used by the models are:

CART: The model accumulates the improvement of the model that occurs when a predictor is used in a split. These values are taken form the rpart object. See rpart::rpart.object().

MARS: MARS models include a backwards elimination feature selection routine that looks at reductions in the generalized cross-validation (GCV) estimate of error. The earth() function tracks the changes in model statistics, such as the GCV, for each predictor and accumulates the reduction in the statistic when each predictor's feature is added to the model. This total reduction is used as the variable importance measure. If a predictor was never used in any of the MARS basis functions in the final model (after pruning), it has an importance value of zero. baguette wraps earth::evimp().

C5.0: C5.0 measures predictor importance by determining the percentage of training set samples that fall into all the terminal nodes after the split. For example, the predictor in the first split automatically has an importance measurement of 100 percent since all samples are affected by this split. Other predictors may be used frequently in splits, but if the terminal nodes cover only a handful of training set samples, the importance scores may be close to zero.

Note that the value column that is the average of the importance scores form each model. The divisor of this average (and the corresponding standard error) is the number of models (as opposed to the number of models that used the predictor). This means that the importance scores for a predictor that was not used in the model has an implicit zero importance.