Home Credit default risk
Change the loan terms and repayment history of a fictional applicant to see how the trained model scores repayment difficulty.
Methods and data
Seven Home Credit tables are joined at applicant level. The model uses payment timing, earlier decisions, monthly balances, and application amounts. Changing a monetary input also recomputes the related ratios. Remaining features come from training-group medians. Missing values follow the tree’s learned rules.
The original project reported a ROC AUC of 0.73. The model bundled with this demo uses a separate training run on 246,008 applications. Three-fold cross-validation within the training partition selects the leaf count. On its 61,503 held-out applications, this model earned a ROC AUC of 0.778. The form uses fictional profiles and runs entirely on your device.
Competition data dictionary ↗