variable_importance {glossa} | R Documentation |
Variable Importance in BART Model
Description
This function computes the variable importance scores for a fitted BART (Bayesian Additive Regression Trees) model using a permutation-based approach. It measures the impact of each predictor variable on the model's performance by permuting the values of that variable and evaluating the change in performance (F-score is the performance metric).
Usage
variable_importance(bart_model, y, x, cutoff = 0, n_repeats = 10, seed = NULL)
Arguments
bart_model |
A BART model object. |
y |
Vector indicating presence (1) or absence (0). |
x |
Dataframe with same number of rows as the length of the vector 'y' with the covariate values. |
cutoff |
A numeric threshold for converting predicted probabilities into presence-absence. |
n_repeats |
An integer indicating the number of times to repeat the permutation for each variable. |
seed |
An optional seed for random number generation. |
Value
A data frame where each column corresponds to a predictor variable, and each row contains the variable importance scores across permutations.