Logistic Regression

Logistic Regression with tidymodels — ISLR Default Dataset

Defaultis the classic textbook binary-classification dataset: 10,000 simulated credit-card customers, predicting whether each one defaults (Yes/No) from their student status, credit card balance, and income.

Packages needed: tidymodels, ISLR (for the Default data)

1. Load and inspect the data

data(Default, package = "ISLR")

default <- as_tibble(Default)

glimpse(default)
Rows: 10,000
Columns: 4
$ default <fct> No, No, No, No, No, No, No, No, No, No, No, No, No, No, No, No…
$ student <fct> No, Yes, No, No, No, Yes, No, Yes, No, No, Yes, Yes, No, No, N…
$ balance <dbl> 729.5265, 817.1804, 1073.5492, 529.2506, 785.6559, 919.5885, 8…
$ income  <dbl> 44361.625, 12106.135, 31767.139, 35704.494, 38463.496, 7491.55…
count(default, default) # note the class imbalance: ~3.3% default = "Yes"

2. Split the data: train / test

set.seed(123)

default_split <- initial_split(default, prop = 0.80, strata = default)
default_train <- training(default_split)
default_test <- testing(default_split)

3. Preprocessing recipe

default_rec <-
  recipe(default ~ student + balance + income, data = default_train) |>
  step_dummy(all_nominal_predictors()) |>
  step_zv(all_predictors()) |>
  step_normalize(all_numeric_predictors())

4. Model specification

logreg_spec <-
  logistic_reg() |>
  set_engine("glm") |>
  set_mode("classification")

5. Workflow: bundle recipe + model

default_wflow <-
  workflow() |>
  add_recipe(default_rec) |>
  add_model(logreg_spec)

6. Fit on the training data

default_fit <- fit(default_wflow, data = default_train)

# Tidy view of coefficients (log-odds scale), sorted by magnitude
default_fit |>
  extract_fit_parsnip() |>
  tidy() |>
  arrange(desc(abs(estimate)))

7. Evaluate with 10-fold cross-validation (on training data)

set.seed(456)
default_folds <- vfold_cv(default_train, v = 10, strata = default)

# "Yes" is the event of interest but comes second alphabetically, so tell
# yardstick which level counts as the "event" for sensitivity/specificity
class_metrics <- metric_set(roc_auc, accuracy, sensitivity, specificity)

default_cv_results <-
  default_wflow |>
  fit_resamples(
    resamples = default_folds,
    metrics   = class_metrics,
    control   = control_resamples(save_pred = TRUE)
  )

collect_metrics(default_cv_results)

8. Final fit on full training set + evaluation on the held-out test set

final_fit <- last_fit(
  default_wflow,
  split   = default_split,
  metrics = class_metrics
)
# Test-set performance
collect_metrics(final_fit)
# Predictions (class + probabilities) on the test set
test_predictions <- collect_predictions(final_fit)
head(test_predictions)

9. Confusion matrix and ROC curve

test_predictions |>
  conf_mat(truth = default, estimate = .pred_class)
          Truth
Prediction   No  Yes
       No  1923   50
       Yes    8   19
test_predictions |>
  roc_curve(truth = default, .pred_Yes) |>
  autoplot() +
  labs(title = "ROC Curve — Predicting Credit Card Default")

10. (Optional) Adjust the classification threshold

Because defaults are rare (~3.3%), the default 0.5 probability threshold will classify almost everyone as “No”. Lowering the threshold trades some precision for much better recall on the minority class.

test_predictions |>
  mutate(.pred_class_alt = factor(if_else(.pred_Yes > 0.2, "Yes", "No"),
    levels = levels(default)
  )) |>
  conf_mat(truth = default, estimate = .pred_class_alt)
          Truth
Prediction   No  Yes
       No  1876   29
       Yes   55   40

11. (Optional) Extract the final fitted workflow for reuse / deployment

final_wflow_fit <- extract_workflow(final_fit)
# saveRDS(final_wflow_fit, "default_logreg_workflow.rds")

# Example: predict on new/unseen data using the finalized workflow
# predict(final_wflow_fit, new_data = some_new_data, type = "prob")
Back to top