Making Predictions Under Lending Regulations
summary
Earnest trains separate classifiers for each decline reason using credit reports plus transaction history, builds each as a random forest (≈500 trees), and stacks their probability outputs into a simple GLM logistic ensemble to make yes/no decisions while producing explicit decline reasons. The Equal Credit Opportunity Act requirement drives this design; nine most-common reasons are modeled, accuracy barely drops versus a monolithic model (AUC ~0.95) and the system assigns valid reasons for over 96% of declined applications.
AI-written summary · about
chapters (13)
- 0:00Introduction to Earnest
- 1:00Earnest’s data sources
- 2:25Automating underwriting
- 4:31Lending regulations overview
- 5:58Equal Credit Opportunity Act
- 7:43Opaque models and decline reasons
- 8:49Traditional risk scorecards
- 9:56Building decline reason models
- 11:19Ensemble methods explained
- 12:20Model feature importances
- 13:46Final ensemble decision
- 15:11Results and transparency gains
- 16:14Q&A session
related talks
39mMachine Learning in Financial Credit Risk Assessment
30mBuilding fairer models for finance
27mData Science and ML with an emphasis on Banking...
38mTo explain or to predict?
47mTrain, Evaluate, Repeat: Building a Credit Card Fraud Detection System
39mTrain, Evaluate, Repeat: Building a Credit Card Fraud Detection System
04h 03mTutorial - Intro to Data Science
42mJust-So Stories for AI: Explaining Black-Box Predictions