Business Context

FinNova, a digital lending platform processing roughly 25,000 personal loan applications per month, wants a simple predictive modeling workflow to estimate whether an applicant will default within 12 months. The goal is to teach core predictive modeling concepts: data preparation, model training, evaluation, and interpretation using Python.

Dataset

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

• Size: 48,000 applications, 30 input features
• Target: default_12m — whether the borrower defaulted within 12 months
• Class balance: 18% default, 82% non-default
• Missing data: 6% missing in employment length, 9% in credit utilization, 3% in annual income

Success Criteria

A good solution should outperform a naive majority-class baseline and produce a model that is easy to explain to risk analysts. Target performance is ROC-AUC >= 0.78 and F1 >= 0.50 on a held-out test set.

Constraints

• Prefer an interpretable baseline before more complex models
• Training should run on a laptop in a few minutes
• Batch scoring is sufficient; no real-time serving requirement
• Avoid data leakage from post-loan information

Deliverables

1. Build a complete binary classification pipeline in Python
2. Explain preprocessing choices for numeric and categorical features
3. Train at least one baseline model and one stronger model
4. Evaluate using appropriate classification metrics and thresholding
5. Summarize key predictive drivers and model tradeoffs