food-ai-nexus/global-food-safety-incidents

---
license: cc-by-4.0
task_categories:
- tabular-classification
- time-series-forecasting
tags:
- food-safety
- food-recalls
- agriculture
- time-series
language:
- en
size_categories:
- 10K<n<100K
pretty_name: Global Food Safety Incidents (FOODAKAI / RASFF)
---

**Global Food Safety Incidents (FOODAKAI / RASFF)** is a tabular time-series dataset containing food safety incident records — border rejections, alerts, recalls, and related notifications — for eight globally traded commodities sourced from the FOODAKAI platform (which aggregates RASFF and other regulatory feeds).

With this dataset, researchers can train machine learning models to forecast the frequency or likelihood of food safety incidents for a given commodity, origin country, and time period.

# Content

- The dataset contains **30,184 incident records** spanning multiple years up to April 2021.
- It covers **8 commodities**: almonds, cereals, chicken, cinnamon, cocoa, sesame seeds, shrimps, and strawberries.
- Each record describes a single food safety incident with its type, hazard, country of origin, and date.
- The `incident_type` column retains the original values from the FOODAKAI/RASFF source. Only formatting artifacts (leading hyphens from data extraction) have been corrected.
- Column names have been standardized to `snake_case` for programmatic access.

# Data Fields

| Column | Type | Description |
| --- | --- | --- |
| `commodity` | string | The food commodity the incident relates to (e.g., `almonds`, `chicken`) |
| `incident_type` | string | The regulatory classification of the incident (see values below) |
| `hazard` | string | The specific hazard identified (e.g., `salmonella`, `aflatoxins`) |
| `hazard_path` | string | The hazard subcategory path in the RASFF taxonomy |
| `origin` | string | Country of origin of the implicated product |
| `date` | string | Date of the incident record in ISO 8601 format (YYYY-MM-DD) |

**Incident Type Values**

The `incident_type` column contains the following values as they appear in the original FOODAKAI/RASFF source. Only leading-hyphen formatting artifacts have been corrected (e.g., `- food` → `food`).

| Value | Global Count | % of Total |
| --- | --- | --- |
| `border rejection` | 19,470 | 64.50% |
| `alert` | 6,487 | 21.49% |
| `food` | 4,138 | 13.71% |
| `feed` | 16 | 0.05% |
| `major food product recall` | 16 | 0.05% |
| `alert notification` | 4 | 0.01% |
| `adulterated food` | 3 | 0.01% |
| `food contact material` | 2 | 0.01% |
| `company's own check` | 2 | 0.01% |
| `poor quality food` | 1 | < 0.01% |
| `unsafe food` | 1 | < 0.01% |

# Uses

The dataset was originally used in the Cornell Food Safety ML Repository to train time-series forecasting models for food safety incidents. It can also be used for anomaly detection, supply chain risk modeling, and food safety surveillance research.

```python
from datasets import load_dataset

# Load all commodities
ds = load_dataset("food-ai-nexus/global-food-safety-incidents")

# Load a single commodity
ds = load_dataset("food-ai-nexus/global-food-safety-incidents", "chicken")
```

# License

This dataset is licensed under the Creative Commons Attribution 4.0 International (CC BY 4.0) License.

# Reference

```bibtex
@misc{cornell_food_safety_ml_2023,
title = {Forecasting Food Safety Incidents},
author = {{Cornell Food Safety ML Repository}},
year = {2023},
howpublished = {\url{https://github.com/Cornell-Food-Safety-ML-Repository/forecasting-food-safety-incidents}},
note = {Data sourced from the FOODAKAI platform (RASFF and other regulatory feeds)}
}
```