---
license: mit
language:
- en
task_categories:
- tabular-classification
- other
tags:
- churn-prediction
- customer-retention
- survival-analysis
- milp
---

# Churn Predict — Data

Companion dataset for the [`churn_predict`](https://github.com/bravo2024/churn_predict) repository
(Customer Churn Prediction & Retention-Budget Optimization).

## Files

| Path | Size | Description |
|---|---|---|
| `data/raw/ordered_open_ecommerce.parquet` | 150 MB | Raw Open E-Commerce 1.0 purchase records (1,761,259 rows, 2018-01 .. 2023-03). Berke et al. 2024. |
| `models/artifacts.pkl` | 197 MB | Fitted churn models, SHAP values, survival fits, and MILP results consumed by the Streamlit dashboard. |

## Why on the Hub

Both files exceed GitHub's 100 MB per-file hard limit, so they are hosted here and fetched at
runtime by `dashboard.py` and `build_dataset.py` (via `huggingface_hub.hf_hub_download`).

## Usage

```python
from huggingface_hub import hf_hub_download

artifacts_path = hf_hub_download(
    repo_id="vivekkopthsd/churn-predict-data",
    filename="models/artifacts.pkl",
)
```

## Source and License

- **Open E-Commerce 1.0** — Berke, A., Calacci, D., Mahari, R., Yabe, T., Larson, K. and Pentland, S.
  (2024). *Open e-commerce 1.0, five years of crowdsourced U.S. Amazon purchase histories with user
  demographics*. Scientific Data, 11. DOI 10.1038/s41597-024-03329-6. MIT (ungated).
- `models/artifacts.pkl` — generated by `train_models.py` (all random seeds 42).
- Ethics: only benign demographics and Amazon-usage survey items; sensitive survey items excluded.