Business Context

NovaBank is hiring a machine learning specialist for its consumer lending team. You are given a realistic credit risk dataset and asked to demonstrate deep expertise in traditional ML by building, explaining, and defending a production-ready default prediction model.

Dataset

The dataset contains historical loan applications and 12-month repayment outcomes from NovaBank's unsecured personal loan product.

• Rows: 240,000 loan applications collected over 30 months
• Target: whether the borrower defaults within 12 months of origination
• Class balance: 11.5% default, 88.5% non-default
• Missing data: 18% missing in bureau features for thin-file applicants; 6% missing in income-related fields

Success Criteria

A strong solution should outperform a regularized logistic regression baseline, achieve robust ranking performance, and provide explanations suitable for model risk review. “Good enough” means ROC-AUC above 0.82, PR-AUC above 0.42, and calibrated probabilities usable for approval policy decisions.

Constraints

• Model must be explainable to risk, compliance, and underwriting teams
• Batch scoring latency must stay under 5 minutes for 100K applications
• Avoid target leakage and justify all preprocessing choices
• Retraining should be feasible monthly with standard Python tooling

Deliverables

1. Build a supervised classification pipeline for default prediction.
2. Compare at least one interpretable baseline and one stronger non-linear model.
3. Explain feature engineering, validation strategy, and leakage prevention.
4. Report business-relevant metrics and threshold recommendations.
5. Describe how you would productionize, monitor, and retrain the model.