Family
XGBoost + Cox Proportional Hazard + LSTM ensemble
in-developmentApache 2.0tabularb2b-facing
Loan Default Predictor.
Loan default probability for underwriting — XGBoost + Cox Proportional Hazard + LSTM ensemble
Foundation · XGBoost + Cox Proportional Hazard + LSTM ensemble
Marketing claim
Credit-risk scores with SHAP-level explainability and Courtroom signatures.Note ·The borrower receives a plain-language explanation of any decision (supporting the DPDP §11 access right); SHAP attributions are operator-facing and are translated into plain-language reasons in the borrower-facing portal.
What it does
The honest description.
Loan default probability for underwriting — XGBoost + Cox Proportional Hazard + LSTM ensemble
Architecture
How it works.
Loan characteristics (amount, tenure, interest rate, disbursal date), payment history (payments_made, days_past_due), and borrower financial features. Multi-horizon default probabilities P(default_30d/60d/90d), traffic-light (Green/Amber/Red) per borrower + ensemble confidence. NOTE: these are the INTENDED outputs of the trained ensemble — nothing currently serves this slug (trainer-only; no backend loader).
Training data & drift
What it learned from, and how we watch it.
Source
Inputs: Loan characteristics (amount, tenure, interest rate, disbursal date), payment history (payments_made, days_past_due), and borrower financial features.
Drift baseline · current quarter
Drift baseline not yet documented for this model.
What it's NOT
- Ensemble: XGBoost (primary) + Cox Proportional Hazard (survival) + LSTM (sequences).
- Time-based train/test split.
- self_healing drift baseline needs no re-base for the PR 6a aux join: aux features are absent-by-default (byte-identical when no account_aggregates upload) and this model is unserved (no live inference-drift stream to compare against), so the TabularStrategy drift reference stays the no-aux distribution.
- Advisory — not a binding lending decision; final adjudication is the partner-lender's responsibility; Mukodi output is advisory.
- Disparate impact metrics across protected attributes (gender / region) will be published in /security#fairness when Phase 7 lands. Until then, partner lenders must independently validate fairness before adverse action.
Who calls this model
Across the house and the API.
Internal routes
No public routes documented yet.
Featured in
/features/credit-riskVersion history · immutable once published
Every release, traceable.
Data Mukodi holds about you is subject to DPDP §12 erasure (use /consent in the app), and any concern about this model can be raised via [email protected] under DPDP §13 grievance, or via /security.