Governed credit-card default early-warning artifacts
This public model repository distributes exact binary artifacts for the
credit-card-default-prediction
portfolio project. The GitHub repository remains the source of validation
authority through reviewed manifests and SHA-256 digests.
Selected release
selected_v1/model.cbm is the unchanged CatBoost winner selected by the
project's four-fit, no-tuning governance workflow. Its intended use is
human-owned outreach prioritisation. It is not approved for adverse action,
autonomous credit decisions, production use, India-specific claims, or
regulatory-compliance claims.
See the GitHub repository for the model card, Release A dossier, subgroup review, final-test evidence, inference contract, registry controls, and limitations.
Dataset
The project does not redistribute training data here. Data is acquired directly from the UCI Default of Credit Card Clients dataset using a checksum-pinned source manifest. The dataset is attributed under CC BY 4.0 and represents a historical 2005 Taiwanese population.
Legacy compatibility files
Files under legacy_v1/ use Python pickle semantics and can execute arbitrary
code. Public hosting does not establish their safety. They must be retrieved
explicitly and loaded only after exact verification against the Git-tracked
legacy trust manifest.
Reproducibility
Consumers retrieve an exact full Hugging Face commit and then independently verify the file SHA-256 values. Normal model loaders remain local-only and never download artifacts implicitly.