Nason/ice-database

---
license: mit
task_categories:
- tabular-classification
- tabular-regression
tags:
- immigration
- ice
- enforcement
- deportation
- detention
- foia
- duckdb
- legal
- policy
- government-data
pretty_name: ICE Enforcement Database
size_categories:
- 10M<n<100M
---

# ICE Enforcement Database

A clean, queryable DuckDB database built from ICE enforcement data published by the [Deportation Data Project](https://law.ucla.edu/academics/centers/center-immigration-law-and-policy/deportation-data-project) (Berkeley Law / UCLA) via FOIA litigation.

**17,824,184 rows** across **5 tables** covering ICE arrests, detainers, detentions, removals, and custody decisions.

Combines two FOIA releases:
- **2023 release** (FY2012-FY2023): arrests, detentions, removals, RCA decisions
- **2025 settlement release** (Sep 2023 - Oct 2025): arrests, detainers, detentions

Every row has a `data_source` column (`release_2023` or `release_2025`) so you can filter by release. Overlapping records are deduplicated, preferring the richer 2025 data.

Built with [ice-database](https://github.com/ian-nason/ice-database).

## Quick Start

### DuckDB CLI

```sql
INSTALL httpfs;
LOAD httpfs;
ATTACH 'https://huggingface.co/datasets/Nason/ice-database/resolve/main/ice.duckdb' AS ice (READ_ONLY);

-- Arrests by month
SELECT DATE_TRUNC('month', apprehension_date) AS month, COUNT(*) AS arrests
FROM ice.arrests
WHERE apprehension_date IS NOT NULL
GROUP BY 1 ORDER BY 1 DESC LIMIT 12;
```

### Python

```python
import duckdb
con = duckdb.connect()
con.sql("INSTALL httpfs; LOAD httpfs;")
con.sql(\"\"\"
ATTACH 'https://huggingface.co/datasets/Nason/ice-database/resolve/main/ice.duckdb'
AS ice (READ_ONLY)
\"\"\")
con.sql("SELECT * FROM ice._metadata").show()
```

DuckDB uses HTTP range requests, so only the pages needed for your query are downloaded.

## Tables

| Table | Description | Rows | Cols | Sources | Date Range |
|-------|-------------|------|------|---------|------------|
| `detentions` | Detention stays (book-in to book-out) | 8,944,408 | 41 | release_2023, release_2025 | 1995-08-31 to 2025-10-16 |
| `rca_decisions` | Release/custody assessment decision history | 3,543,467 | 42 | release_2023 | |
| `removals` | Deportation/removal records | 2,771,219 | 29 | release_2023 | 0212-06-29 to 2023-10-27 |
| `arrests` | ICE administrative arrests | 2,168,784 | 23 | release_2023, release_2025 | 2011-10-01 to 2025-10-16 |
| `detainers` | Detainer requests issued to jails/prisons | 396,306 | 63 | release_2025 | 1989-09-25 to 2025-10-15 |

## Key Features

### Linked Records
Tables share a `unique_id` field for tracing individuals across the enforcement pipeline: arrests -> detainers -> detentions -> removals.

### Pre-built Views
- `v_arrest_to_detention` - Arrests joined to detention stays
- `v_enforcement_pipeline` - Full pipeline: arrest -> detention -> removal
- `v_daily_arrests` - Daily arrest counts by data source

### Multi-release Deduplication
Where both releases cover the same period, records are deduplicated on key fields (unique_id + date + facility) with the richer 2025 release preferred.

## Data Source

[Deportation Data Project](https://law.ucla.edu/academics/centers/center-immigration-law-and-policy/deportation-data-project) (Berkeley Law / UCLA). Data obtained through FOIA litigation against ICE.

## License

Database build code: MIT. Underlying data: public domain (U.S. government records released via FOIA).

## GitHub

Full source code, build instructions, and example queries: [github.com/ian-nason/ice-database](https://github.com/ian-nason/ice-database)