Decide which loan applicants are bad credit risks when only 30% of them are, and when missing a bad applicant costs five times more than wrongly rejecting a good one. Build a ColumnTransformer + LogisticRegression pipeline, compare it with a class-weighted version using ROC and precision-recall curves, tune the decision threshold against the business cost matrix, and check whether the predicted probabilities are calibrated. The dataset is fetched from OpenML on first run.
Data Download
Exploration & Preprocessing
Training
Evaluation / Metrics
Premium Project
This guided, GPU-powered project is available to Premium members. Upgrade to train models end-to-end in an interactive notebook.