subtopic

Rule-based risk tiers (a multi-class label of our own)

Create a 4-tier label A/B/C/D (instead of only Good/Bad) from 2–3 strong drivers, using bad rates from the TRAIN split. Concepts: Risk tiers; rule-based segmentation; multi-class target design Project A: Example rule: tier from checking status × duration band × credit history (e.g. D = overdrawn AND duration > 24 months; A = no checking account AND paid duly). Verify bad rate increases A → D on train and test. Project B: Not applicable.