⚽ FIFA World Cup 2026 Predictor

A data-driven football prediction project for the upcoming FIFA World Cup 2026, combining historical match data, team form signals, and an XGBoost-based model to estimate match outcomes and tournament title chances.

🌍 Project Overview

This project is designed to help fans, analysts, and football enthusiasts explore how a modern machine learning pipeline can be used to:

• predict match outcomes for international football fixtures
• estimate win / draw / loss probabilities
• simulate a World Cup 2026 tournament and generate provisional title odds
• visualize predictions through a lightweight Streamlit dashboard

🏆 What the project includes

• historical football results ingestion
• leak-free feature engineering for pre-match prediction
• dynamic Elo ratings and rolling form indicators
• training of a multiclass XGBoost model
• match probability predictions for any pair of teams
• a tournament simulation for provisional 2026 title odds
• an interactive dashboard for quick exploration

📊 Data Source

The project uses historical international football results from the public dataset martj42/international_results, mirrored and prepared for model training.

🧠 Model Approach

The model predicts outcomes from the perspective of Team A:

• 0 = Team A loss
• 1 = draw
• 2 = Team A win

Core features include:

• dynamic Elo ratings for both teams
• Elo difference
• rolling points over the last 5 matches
• rolling goals scored / conceded
• goal difference
• matches played
• neutral venue flag
• home advantage signal
• tournament importance
• expected score estimates from Elo
• rest days

🏗️ Project Structure

• scripts/download_data.py downloads raw match data
• scripts/build_features.py creates training-ready pre-match features
• scripts/train_match_model.py trains the predictive model
• scripts/predict_worldcup_2026.py runs the tournament simulation
• src/features/build_match_features.py handles feature engineering logic
• src/models/predict_match.py provides match prediction utilities
• src/simulation/worldcup_simulator.py simulates the World Cup bracket
• dashboard_app.py launches the Streamlit web app

🚀 Getting Started

1. Install dependencies

pip install -r requirements.txt

2. Run the full pipeline

python scripts/run_pipeline.py

3. Launch the dashboard

streamlit run dashboard_app.py

⚙️ Individual Scripts

python scripts/download_data.py
python scripts/build_features.py
python scripts/train_match_model.py
python scripts/predict_worldcup_2026.py

🔎 Note on 2026 Title Odds

The tournament simulation is currently a strong prototype rather than an official FIFA forecast. It uses:

• the official 48-team group structure
• placeholder handling for unresolved playoff matchups
• a general seeded knockout structure rather than a fully exact tournament bracket

The match prediction model itself is the core reliable component, while the title probabilities provide an exciting preview of how the tournament could unfold.

🔮 Future Improvements

Potential next steps include:

• replacing the provisional bracket logic with the exact FIFA World Cup 2026 draw structure
• adding official FIFA rankings or external Elo snapshots as features
• tuning model hyperparameters and improving probability calibration
• exploring scoreline prediction models such as Poisson or bivariate Poisson
• deploying the app for public access