This function generates Precision-Recall and ROC curves for sample subgroups, facilitating fairness analysis of a binary classification model.
Arguments
- task
mlr3 binary classification task object specifying the task details
- trained_model
mlr3 trained learner (model) object obtained after training
- splits
mlr3 object defining data splits for train and test sets
- target_variable
character, the variable from the dataset used to test the model's performance against
- var_levels
list, defining the levels for the specified variable
Examples
library("explainer")
seed <- 246
set.seed(seed)
# Load necessary packages
if (!requireNamespace("mlbench", quietly = TRUE)) stop("mlbench not installed.")
if (!requireNamespace("mlr3learners", quietly = TRUE)) stop("mlr3learners not installed.")
if (!requireNamespace("ranger", quietly = TRUE)) stop("ranger not installed.")
# Load BreastCancer dataset
utils::data("BreastCancer", package = "mlbench")
target_col <- "Class"
positive_class <- "malignant"
mydata <- BreastCancer[, -1]
mydata <- na.omit(mydata)
sex <- sample(
c("Male", "Female"),
size = nrow(mydata),
replace = TRUE
)
mydata$age <- as.numeric(sample(
seq(18, 60),
size = nrow(mydata),
replace = TRUE
))
mydata$sex <- factor(
sex,
levels = c("Male", "Female"),
labels = c(1, 0)
)
maintask <- mlr3::TaskClassif$new(
id = "my_classification_task",
backend = mydata,
target = target_col,
positive = positive_class
)
splits <- mlr3::partition(maintask)
mylrn <- mlr3::lrn(
"classif.ranger",
predict_type = "prob"
)
mylrn$train(maintask, splits$train)
# sex is chosen for fairness analysis
Fairness_results <- eFairness(
task = maintask,
trained_model = mylrn,
splits = splits,
target_variable = "sex",
var_levels = c("Male", "Female")
)
#> Warning: D not labeled 0/1, assuming benign = 0 and malignant = 1!
#> Warning: D not labeled 0/1, assuming benign = 0 and malignant = 1!
#> Warning: D not labeled 0/1, assuming benign = 0 and malignant = 1!