Business Context

BrightLend, a mid-size consumer lending platform, wants a baseline predictive model to estimate whether a loan applicant will default within 12 months of origination. The goal is to help analysts demonstrate sound predictive modeling fundamentals: problem framing, feature preparation, model training, and evaluation.

Dataset

You are given a historical loan application dataset with one row per funded loan.

• Size: 48K funded loans, 35 input features
• Target: default_12m — whether the borrower defaulted within 12 months
• Class balance: 14% default, 86% non-default
• Missing data: 10% missing in employment-related fields, 6% missing in credit bureau attributes

Success Criteria

A good solution should outperform a naive majority-class baseline and produce a model that is understandable enough for risk analysts to review. Target performance is ROC-AUC >= 0.78 and F1 >= 0.45 on a held-out test set, with clear discussion of threshold selection.

Constraints

• Prefer interpretable or semi-interpretable models suitable for analyst review
• Batch scoring is sufficient; no real-time serving requirement
• Training should run on a standard laptop in minutes
• Avoid leakage from post-loan features

Deliverables

1. Build a baseline classification pipeline for default_12m
2. Explain feature preprocessing choices for numeric and categorical columns
3. Compare at least two models and justify the final selection
4. Evaluate using appropriate classification metrics beyond accuracy
5. Identify the most important predictors and discuss practical tradeoffs