Plot Variable Importance from a caretStack Model

This function plots the variable importance from a stacked ensemble model (`caretStack`), separating original features from new (engineered) features. It optionally includes cross-group summary statistics (mean, sum, or max) from one feature group into the other for visual reference. It returns a ggplot object; if both original and new features are present, the plot will contain two facets (original and new features) within the same figure. This is useful for diagnosing which group of features contributes more to the stacked model.

Usage

plot_variable_importance(stack_model, newdata, stat_type = NULL)

Arguments

stack_model

A caretStack model trained using caretEnsemble. It should have an attribute original_features listing the original input variables. If this attribute is missing, all variables are treated as "new".

newdata

A data frame containing the data used for calculating variable importance. Typically this should be the validation or test set.

stat_type

Optional character string indicating which summary statistic of the opposite group to include as a gray bar for reference. Must be one of "mean", "sum", or "max". If NULL, no statistic is shown. If invalid, an error is thrown.

Value

A ggplot object. If the model includes both original and new features, the plot will contain two facets ("Original Features" and "New Features"). If stat_type is provided, a gray bar appears in each plot representing the selected summary statistic from the opposite group (e.g., mean of new features shown in the original features plot).

Details

- Variable importance is computed using caret::varImp. - If the model lacks the original_features attribute, all variables are considered new. - Requires the packages: data.table, ggplot2, and caret.