A machine learning system that uses XGBoost to predict flight risk scores (0-100) based on weather conditions for multiple US geographic zones. The system analyzes METAR weather data and airport observations integrated with actual flight delay data to assess weather-related flight risks across 5 major US regions.

• Real-Time METAR Integration: Fetch current weather data directly from NOAA Aviation Weather API
• Automated Data Processing: Parses METAR format weather data and combines it with airport observations
• Machine Learning Model: XGBoost-based regression model for accurate risk prediction
• 0-100 Risk Scoring: Simple scoring system where 100 = maximum flight risk
• Multi-Zone Support: 5 regional models covering all US airports
• Comprehensive Feature Engineering: Uses multiple weather parameters including visibility, ceiling, wind speed, and more
• 34+ Airport Support: Loads airports from Excel file for easy expansion

1. Create and activate a virtual environment:

python -m venv myenv
myenv\Scripts\activate  # Windows
source myenv/bin/activate  # Linux/Mac

2. Install required packages:

pip install pandas numpy xgboost scikit-learn joblib

The system processes two main data sources:

• METAR Data (4152601.csv): Standard weather observations in METAR format
• Airport Data (AUS.csv): Austin Bergstrom International Airport weather observations

1. Parse METAR format fields (temperature, wind, visibility, ceiling, pressure)
2. Convert units and handle missing values
3. Combine datasets by date
4. Engineer additional features (temp spread, gust factor, ceiling/visibility ratio, etc.)

python Src\data_processing.py

This creates Data/cleaned_weather_data.csv with processed data.

The model uses the following key features:

• Temperature (°C)
• Dewpoint (°C)
• Wind direction and speed (knots)
• Visibility (km)
• Ceiling height (feet)
• Sea level pressure (mb)
• Wind gusts
• Relative humidity

• Temperature spread (temp - dewpoint)
• Gust factor (gust / wind speed)
• Ceiling to visibility ratio
• Pressure changes
• Time features (hour, month, day/night)

The baseline risk score is calculated from:

• Visibility (0-40 points): Lower visibility = higher risk
• Ceiling (0-30 points): Lower ceiling = higher risk
• Wind Speed (0-20 points): Higher winds = higher risk
• Wind Gusts (0-10 points): Strong gusts = higher risk

python Src\model_training.py

The model will be saved to models/ directory:

• xgboost_flight_risk_model.pkl - Trained model
• scaler.pkl - Feature scaler
• feature_names.txt - List of features

The system includes 5 regional models trained on actual flight delay data:

Note on R² scores: Weather explains a modest fraction of flight delays because many delays stem from non-weather causes (ATC congestion, maintenance, cascading delays). An R² around 0.1–0.3 is expected for weather-focused flight risk models. The Southeast model captures the strongest weather signal.

from Src.predict_risk import predict_risk, interpret_risk_score

# Define weather conditions
weather = {
    'temperature_c_aus': 15.0,
    'dewpoint_c_aus': 10.0,
    'wind_direction_aus': 270,
    'wind_speed_kts_aus': 5.0,
    'visibility_km_aus': 10.0,
    'ceiling_ft_aus': 5000.0,
    'sea_level_pressure_mb_aus': 1013.0,
    'gust': 0.0,
    'relh': 70.0,
    'temp_spread': 5.0,
    'gust_factor': 0.0,
    'ceiling_vis_ratio': 1.5,
    'pressure_change': 0.0,
    'humidity': 70.0,
    'hour': 12,
    'month': 6,
    'is_night': 0
}

# Predict risk score
risk_score = predict_risk(weather)
category = interpret_risk_score(risk_score[0])

print(f"Risk Score: {risk_score[0]:.1f}/100")
print(f"Category: {category}")

python Src\predict_risk.py

This runs example predictions with sample weather conditions.

To predict risk for a specific US region:

from Src.predict_zone_risk import predict_zone_risk

# Define weather conditions for a specific zone
weather = {
    'temperature_c': 20.0,
    'dewpoint_c': 15.0,
    'wind_direction': 270,
    'wind_speed_kts': 15.0,
    # ... all required features
}

# Predict for Northeast
risk_score = predict_zone_risk('Northeast', weather)

Available zones: Northeast, PacificCoast, RockyMountains, CentralPlains, Southeast

Option 1: Interactive Interface with Real-Time METAR

To test with real-time weather data or custom values:

python interactive_test.py

Features:

• Real-time METAR fetching from NOAA API
• Support for 34+ airports from Excel file
• Automatic zone detection and model selection
• Manual weather entry option
• Detailed risk interpretation

Quick Example:

Enter Airport Code: AUS
Fetch current weather from NOAA METAR? (y/n): y
Fetches current weather automatically!

Option 2: Legacy Interactive Interface

python Src\test_model.py

This will prompt you to enter weather conditions and provide real-time risk predictions. You can:

• Enter your own values or press Enter to use defaults
• Test multiple scenarios in one session
• See detailed risk breakdowns and recommendations

Option 2: Programmatic Testing

Edit and run the example test file to test with your own custom values:

python Src\example_test.py

Edit the weather dictionaries in Src/example_test.py to customize the test conditions. This is useful for:

• Batch testing multiple scenarios
• Programmatic integration
• Reproducible test cases

• 0-20: Very Low Risk
• 20-40: Low Risk
• 40-60: Moderate Risk
• 60-80: High Risk
• 80-100: Very High Risk

• Visibility: 10 km
• Ceiling: 5000 ft
• Wind Speed: 5 knots
• Risk Score: 4.7 (Very Low Risk)

• Visibility: 0.5 km
• Ceiling: 200 ft
• Wind Speed: 35 knots
• Gusts: 45 knots
• Risk Score: 99.7 (Very High Risk)

WeatherScores/
├── Data/
│   ├── 4152601.csv          # METAR weather data
│   ├── AUS.csv              # Airport weather data
│   └── cleaned_weather_data.csv  # Processed data
├── Src/
│   ├── data_processing.py   # Data cleaning and processing
│   ├── model_training.py    # Model training
│   └── predict_risk.py      # Risk prediction
├── models/                   # Saved models
│   ├── xgboost_flight_risk_model.pkl
│   ├── scaler.pkl
│   └── feature_names.txt
└── README.md

• Python 3.7+
• pandas
• numpy
• xgboost
• scikit-learn
• joblib

This project is for educational and research purposes.

• The model was trained on Austin Bergstrom International Airport weather data
• Risk scores are estimates based on historical weather patterns
• Always consult with aviation professionals for actual flight decisions
• Weather conditions can change rapidly; real-time monitoring is essential for flight safety