Datasets:

cjc0013
/

Ufo_data_clustered

datasets:
  - kaggle
language:
  - en
task_categories:
  - text-classification
  - text-retrieval
  - time-series-forecasting
pretty_name: UFO Sightings Unified Semantic Dataset
size_categories:
  - 100K<n<1M
license: mit
tags:
  - ufo
  - ufology
  - anomalous-phenomena
  - embeddings
  - hdbscan
  - semantic-search
  - cluster-analysis
  - dataset
  - geospatial
  - moon-illumination
  - aviation

UFO Sightings – Cleaned & Unified Dataset (~327k rows)

This dataset merges several publicly available UFO sighting datasets from Kaggle into one cleaned, standardized, and enriched file. The goal is simply to provide a consolidated dataset instead of many fragmented sources with inconsistent formatting.

This release contains a single JSONL file with approximately 327,000 records.

No private or identifying information was present in the original data.

📦 Source

All entries originate from publicly available UFO sighting datasets on Kaggle. Each row corresponds to a single reported sighting. Source Files located in source folder

🧹 Cleaning / Normalization Performed

All rows in this unified file were standardized using the same basic rules:

timestamps parsed and converted into a consistent t_utc (ISO-8601, UTC)
city/state/country fields harmonized where possible
latitude/longitude coerced to floats
basic HTML/unicode cleanup in free-text descriptions (text)
invalid or fully unparseable rows removed
source field preserved as src

No interpretation or filtering based on content was performed.

✨ Added Contextual Fields

A small number of lightweight “sidecar” fields were added based on timestamp + coordinates:

moon_illum — moon illumination fraction
moon_alt_deg — moon altitude in degrees
nearest_airport_code — closest airport (ICAO)
nearest_airport_km — distance to that airport in km
wx_bucket — rough weather bucket (coarse category)

These values are approximate and should be treated as exploratory metadata only.

🧩 Clustering Fields (Included in the File)

The dataset includes two fields that come from text-similarity grouping:

cluster_id — numeric label
prob — membership confidence

These reflect text similarity, not verified categories or event types. They are included because they were already part of the cleaned file.

📝 Field Reference

Each row has the following structure (example):

{
  "uid": "scrubbed/row327047",
  "t_utc": "2013-09-09T09:51:00.000Z",
  "lat": 32.7152778,
  "lon": -117.1563889,
  "text": "2 white lights zig-zag over Qualcomm Stadium...",
  "src": "scrubbed",
  "city": "san diego",
  "state": "ca",
  "country": "US",

  "cluster_id": 725,
  "prob": 1.0,

  "moon_illum": 0.163603127,
  "moon_alt_deg": -67.0003509521,
  "nearest_airport_km": 3.7174715996,
  "nearest_airport_code": "KSAN",
  "reports_z": null,
  "wx_bucket": "unknown"
}

Field descriptions

Field	Type	Notes
`uid`	string	Stable row identifier
`t_utc`	string	Event timestamp, ISO-8601 UTC
`lat`, `lon`	float	Approximate coordinates
`city`, `state`, `country`	string	Cleaned location fields (best-effort)
`text`	string	Free-text sighting description
`src`	string	Original Kaggle dataset source
`cluster_id`	int	Text-similarity cluster (for research use only)
`prob`	float	Cluster membership probability
`moon_illum`	float	Moon illumination (0–1)
`moon_alt_deg`	float	Moon altitude in degrees
`nearest_airport_km`	float	Distance to nearest airport
`nearest_airport_code`	string	ICAO code
`wx_bucket`	string	Approximate weather category
`reports_z`	float/null	Unused placeholder field (kept for completeness)

⚠️ Notes & Limitations

Accuracy of timestamps and locations depends entirely on original reporting.
Weather buckets are coarse (not NOAA-grade).
Airport distances are approximate nearest-neighbor lookups.
Cluster labels are based solely on text similarity and do not reflect event reality.
No claims are made about the nature or validity of any sighting.

📄 License

Source data was public on Kaggle. This cleaned, merged, and lightly enriched version is released for research and educational use. Users should follow the original dataset licensing terms.