gafferOSv2 ⚽

A tactical intelligence system for semi-professional football teams, combining match data and machine learning to generate data-driven insights and decision support.

---

🚀 Overview

gafferOSv2 uses historical match data (StatsBomb) and engineered performance metrics to:

• Predict match outcomes
• Analyze team strengths (attack vs defense dynamics)
• Model team form and matchup differences
• Lay the foundation for tactical decision-making (e.g., formation adjustments)

---

⚙️ Current Progress

• PostgreSQL database and REST API implemented
• 1100+ matches processed (La Liga, UCL, Premier League)
• 30+ engineered features from match data
• Rolling form and team-vs-opponent differential features
• Exploratory Data Analysis completed
• Initial XGBoost model trained (WIP, ~74% accuracy)

---

📊 Key Insights

• Match outcomes are primarily driven by:

  • Offensive output
  • Defensive solidity
  • Shot quality

• Home advantage has a measurable impact

• Several features are highly correlated, requiring pruning

• Teams exhibit strong tactical identity:

  • ~63% of matches use their most common formation

---

🧠 Core Idea

Rather than asking:

“Which formation is best?”

gafferOS aims to answer:

“What is the best tactical adjustment for this team in this situation?”

This is achieved by:

• Learning team tendencies (form + formation identity)
• Comparing realistic tactical variations
• Evaluating outcomes using the trained model

---

🛠️ Tech Stack

Python • FastAPI • PostgreSQL • XGBoost

---

🔄 Pipeline

Data → Feature Engineering → ML Model → API → Tactical Insights

---

📁 Structure

backend/
├── api/        # REST endpoints
├── core/       # business logic
├── db/         # database models
├── ml/         # data + training pipeline
└── main.py

---

🎯 Goal

To make practical tactical analytics accessible to teams without elite-level resources.

---

📄 Status

Active development — evolving from prediction system to tactical decision engine