An Expected Goals (xG) model that predicts the probability of scoring based on StatsBomb data. The project uses machine learning techniques to analyze factors most influencing shot effectiveness in football. Applied models (Logistic Regression, Random Forest, XGBoost) combined with Beta calibration technique create a highly accurate predictive tool. Analysis results confirm the crucial role of shot geometry and defender influence on goal-scoring probability.

Expected Goals (xG) is one of the most important metrics used in modern football analytics. It allows for evaluating shot quality regardless of whether they resulted in a goal. In this project, I built my own xG model to better understand factors affecting shot effectiveness and create a tool that can be used for match analysis and player evaluation.

The data used comes from StatsBomb’s open dataset from the 2015/2016 season for five top European leagues:

• Premier League (England)
• La Liga (Spain)
• Bundesliga (Germany)
• Serie A (Italy)
• Ligue 1 (France)

The data contains detailed information about each shot, including position on the pitch, shot type, circumstances of the shot, and positioning of other players at the moment of the shot.

Note: The repository does not include data files by default. You need to run the data_collector.ipynb notebook first to download the data from StatsBomb’s open dataset.

https://github.com/statsbomb/open-data

• Extraction of relevant shot-related variables
• Transformation of raw location data into useful geometric features
• Categorization of shot types and body parts used for shots

• Geometric: shot angle, distance from goal
• Contextual: number of defenders on shot line, goalkeeper presence
• Technical: dominant vs non-dominant foot shots, first-time shots
• Situational: shots under pressure, shots after dribbling

Testing and comparison of three algorithms:

1. Logistic Regression
2. Random Forest
3. XGBoost

Application of Beta Calibration technique to calibrate probabilities, which significantly improved model prediction quality.

1. Shot geometry is crucial – shot angle and distance from goal are the strongest predictors
2. Defenders on shot line – each additional defender significantly decreases goal-scoring probability
3. First-time shots have higher effectiveness than those preceded by ball control
4. Model calibration is crucial – all models before calibration significantly overestimated probabilities

• Language: Python 3.7+
• Data Analysis: Pandas, NumPy
• ML Models: Scikit-learn, XGBoost
• Visualization: Matplotlib, Seaborn, Mplsoccer
• Data Source: StatsBombPy

Football-xG-Predictor/
├── notebooks/                 
│   ├── data_collection.py      # Data collection script
│   └── xg_model.ipynb          # Main notebook with xG model
├── src/                        
│   ├── __init__.py             # Package initialization file
│   ├── preprocessing.py        # Data preprocessing functions
│   ├── feature_engineering.py  # Feature engineering
│   ├── modeling.py             # Model implementation
│   ├── evaluation.py           # Metrics and model evaluation
│   └── visualization.py        # Visualizations
├── data/                       # Data folder (created after running data_collector.ipynb)
├── assets/                     # Graphics and visualizations
├── requirements.txt            # Dependencies
└── README.md                   # Project description / this file

1. Clone the repository:

git clone https://github.com/bsobkowicz1096/Football-xG-Predictor.git

2. Navigate to the project directory:

cd Football-xG-Predictor

3. Create a virtual environment (optional but recommended):

python -m venv venv
source venv/bin/activate  # On Linux/macOS
venv\Scripts\activate     # On Windows

4. Install dependencies:

pip install -r requirements.txt

5. Run the notebook:

jupyter notebook notebooks/football_xg_predictor.ipynb

Note: The project uses publicly available StatsBomb data, used in accordance with their license terms.