---
license: other
task_categories:
  - tabular-classification
tags:
  - credit-risk
  - binary-classification
  - finance
size_categories:
  - 100K<n<1M
---

# Home Credit - Credit Risk Model Stability (Processed)

Processed subset of the [Home Credit - Credit Risk Model Stability](https://www.kaggle.com/competitions/home-credit-credit-risk-model-stability) Kaggle competition dataset, prepared for the RWDS article on information-theoretic foundations of WoE, IV, and PSI.

## Dataset

- **Rows**: 522,596 loan applications
- **Columns**: 48 (32 predictors + target + metadata + 11 applicant/CB attributes)
- **Source tables**: `train_base`, `train_static_0_1` (depth=0, internal), `train_static_cb_0` (depth=0, external), `train_person_1` (depth=1, internal)
- **Target**: binary default indicator (1 = default, 0 = non-default)

## Column Reference

### Special Columns

| Column | Type | Description |
|--------|------|-------------|
| `case_id` | int | Unique identifier for each credit case |
| `date_decision` | datetime | Date when the loan decision was made |
| `MONTH` | datetime | Month of decision (for aggregation) |
| `WEEK_NUM` | int | Week number (for temporal stability / PSI analysis) |
| `target` | int | 1 = client defaulted, 0 = non-default |

### Applicant Attributes (from `person_1`)

| Column | Type | Coverage | Description |
|--------|------|----------|-------------|
| `sex_738L` | str | 100% | Applicant sex (F / M) |
| `age` | float | 100% | Age in years (derived from `birth_259D`, reference 2024-01-01) |
| `education_927M` | str | 100% | Education level (6 masked categories) |
| `incometype_1044T` | str | 100% | Income type: PRIVATE_SECTOR_EMPLOYEE, SALARIED_GOVT, RETIRED_PENSIONER, EMPLOYED, SELFEMPLOYED, OTHER, HANDICAPPED_2, HANDICAPPED_3 |
| `familystate_447L` | str | 46% | Marital status: MARRIED, SINGLE, WIDOWED, DIVORCED, LIVING_WITH_PARTNER |
| `empl_employedtotal_800L` | str | 23% | Employment tenure: LESS_ONE, MORE_ONE, MORE_FIVE |
| `language1_981M` | str | 100% | Language group (3 masked categories) |
| `mainoccupationinc_384A` | float | 100% | Main occupation income amount |

### Credit Bureau Attributes (from `static_cb_0`)

| Column | Type | Coverage | Description |
|--------|------|----------|-------------|
| `maritalst_385M` | str | 100% | Marital status from credit bureau (6 masked categories) |
| `requesttype_4525192L` | str | 57% | Request type: DEDUCTION_6, PENSION_6, SOCIAL_6 |
| `description_5085714M` | str | 100% | Product description (2 masked categories) |

### Predictor Suffix Codes

Column names end with a capital letter denoting the transformation group applied by the data provider:

| Suffix | Transformation |
|--------|---------------|
| `P` | Days Past Due (DPD) transform |
| `M` | Masking categories |
| `A` | Amount transform |
| `D` | Date transform |
| `T` | Unspecified transform |
| `L` | Unspecified transform |

For example, `maxdbddpdtollast6m_4187119P` is a DPD-transformed feature.

### Feature Groups

**Days Past Due (5 features)**

| Feature | Description |
|---------|-------------|
| `maxdpdfrom6mto36m_3546853P` | Max DPD from 6 to 36 months |
| `maxdpdlast12m_727P` | Max DPD last 12 months |
| `maxdpdlast24m_143P` | Max DPD last 24 months |
| `maxdpdlast3m_392P` | Max DPD last 3 months |
| `maxdpdlast6m_474P` | Max DPD last 6 months |

**Credit Bureau (6 features)**

| Feature | Description |
|---------|-------------|
| `numberofqueries_373L` | Number of credit enquiries |
| `days120_123L` | CB 120-day indicator |
| `days180_256L` | CB 180-day indicator |
| `days30_165L` | CB 30-day indicator |
| `days90_310L` | CB 90-day indicator |
| `days360_512L` | CB 360-day indicator |

**User Profile (21 features)**

| Feature | Description |
|---------|-------------|
| `numinstls_657L` | Number of instalments |
| `numinstlsallpaid_934L` | Instalments all paid |
| `pctinstlsallpaidlat10d_839L` | % instalments paid late (10d) |
| `totalsettled_863A` | Total settled amount |
| `totaldebt_9A` | Total debt |
| `currdebt_22A` | Current debt |
| `credamount_770A` | Credit amount |
| `mobilephncnt_593L` | Mobile phone count |
| `homephncnt_628L` | Home phone count |
| `numactivecreds_622L` | Number of active credits |
| `applicationcnt_361L` | Application count |
| `applications30d_658L` | Applications last 30 days |
| `applicationscnt_1086L` | Applications count |
| `applicationscnt_464L` | Applications count |
| `applicationscnt_629L` | Applications count |
| `avgdbddpdlast24m_3658932P` | Avg DPD last 24 months |
| `amtinstpaidbefduel24m_4187115A` | Amount paid before due (24m) |
| `maxdbddpdlast1m_3658939P` | Max DPD last 1 month |
| `maxdbddpdtollast12m_3658940P` | Max DPD tolerance last 12m |
| `maxdbddpdtollast6m_4187119P` | Max DPD tolerance last 6m |
| `numinstpaidlate1d_3546852L` | Instalments paid late (1d) |

## Depth Levels (Original Dataset)

This processed file uses only depth=0 tables (static features tied directly to `case_id`) plus primary applicant demographics from depth=1. The full competition dataset includes:

- **depth=0**: Static features directly tied to a `case_id`
- **depth=1**: Historical records indexed by `num_group1` (requires aggregation)
- **depth=2**: Historical records indexed by `num_group1` and `num_group2`

When `num_group1=0`, it refers to the applicant (the person who applied for the loan).

## Citation

```bibtex
@misc{homecredit2024,
  author = {Daniel Herman and Tomas Jelinek and Walter Reade and Maggie Demkin and Addison Howard},
  title = {Home Credit - Credit Risk Model Stability},
  year = {2024},
  howpublished = {\url{https://kaggle.com/competitions/home-credit-credit-risk-model-stability}},
  note = {Kaggle}
}
```

## License

This dataset is subject to the [Kaggle competition rules](https://www.kaggle.com/competitions/home-credit-credit-risk-model-stability/rules). Please review the terms before use.