---
language:
- en
license: mit
tags:
- aviation
- apu
- turnaround
- airport-operations
- decarbonization
- green-aviation
size_categories:
- 1M<n<10M
task_categories:
- tabular-regression
- tabular-classification
pretty_name: APU-Guard Aviation Turnaround & Emissions Benchmark Dataset
---

# APU-Guard Dataset Benchmark

This dataset repository hosts the multi-source raw and processed benchmark datasets used by **APU-Guard** (Autonomous APU Optimization & Ground Utility Oversight System).

## Repository Overview

- **Repository**: `dimoontri/apu-guard-data`
- **Total Size**: ~6.8 GB
- **Project**: APU-Guard (Decarbonizing Aircraft Auxiliary Power Unit operations during turnaround)

## Dataset Structure

```
├── raw/
│   ├── turnaround_20_stands/                     # Mendeley 20-Stand Ground Truth (7,122 turnarounds)
│   ├── turnaround_20_stands.csv
│   ├── De-identified 20-Stand Aircraft Turnaround Dataset.zip
│   ├── flight_fuel_consumption/                  # NASA DASHlink Continuous Flight Trajectory (DataSet1_Q, ~1.67 GB)
│   ├── fq2/                                      # FlightQuest 2 Aggregate Operational / METAR (~3.99 GB)
│   │   ├── FQ2DataRelease2_AggregateTrain.7z
│   │   ├── FlightStatsDataDictionary-Rev2.xlsx
│   │   ├── training3_flighthistory_rev1.zip
│   │   └── training3_metar_reports_rev1.zip
│   ├── jetsetgo/                                 # JetSetGo Turnaround Sensor Telemetry (~605 MB)
│   ├── cax/                                      # CAX Complex Turnaround Logs (~459 MB)
│   ├── weather/                                  # LEVX Metar / Surface Weather (~46 MB)
│   ├── gse/                                      # NREL eGSE & Airport Ground Equipment (~37 MB)
│   ├── openalaqs/                                # OpenALAQS Reference Emission Profiles
│   ├── airports.parquet                          # Global Airport Stand / Gate Lookups
│   ├── flightlist_train.parquet                  # OpenSky Network ADS-B Turnaround Logs
│   └── fuel_train.parquet                        # Fuel burn reference baselines
└── processed/
    └── apu_guard_training.parquet                # Stage 1 Normalized Master Dataset (7,122 turnarounds, 33 columns)
```

## Data Provenance & Contract (v2.1.0)

| Source | Role | Primary Features | Provenance |
| --- | --- | --- | --- |
| **Mendeley 20-Stand** | Primary Turnaround Truth | `apu_minutes`, `on_block`, `off_block`, `stand`, `charger_kwh_linked` | Mendeley Data (CC BY 4.0) |
| **NASA DASHlink** | Reference Fuel Burn | `DataSet1_Q` continuous telemetry, APU fuel ~108 kg/h | NASA DASHlink Public Domain |
| **FlightQuest 2** | Operational Context | METAR weather, aggregate turnaround schedules | GE / Kaggle FlightQuest |
| **JetSetGo** | Turnaround Context | Complex multi-sensor turnaround logs | Open Aviation Data |
| **LEVX Weather** | Surface Weather | Temperature (°C), Dew Point, Wind Speed, QNH | LEVX Airport Surface Stations |
| **NREL eGSE** | Ground Fleet Context | Hourly ground power demand distributions | NREL Open Energy Data |
| **OpenALAQS / ICAO** | Reference Profiles | APU baseline fuel flow & NOx emission indices | ICAO Engine Databank |

## Usage with Python & Hugging Face

### Download entire dataset:
```python
from huggingface_hub import snapshot_download

local_dir = snapshot_download(
    repo_id="dimoontri/apu-guard-data",
    repo_type="dataset",
    local_dir="data"
)
print(f"Downloaded to {local_dir}")
```

### Direct Parquet Loading with DuckDB / Pandas:
```python
import pandas as pd

# Load Stage 1 Master Parquet directly from Hugging Face
url = "https://huggingface.co/datasets/dimoontri/apu-guard-data/resolve/main/processed/apu_guard_training.parquet"
df = pd.read_parquet(url)
print(f"Loaded {len(df):,} turnarounds. Columns: {list(df.columns)}")
```