⚽ FUTSCOUT: Position-Aware Deep Learning System for Intelligent Football Player Scouting

Live Demo: FUTSCOUT Web App

FUTSCOUT is an intelligent football analytics platform designed to go beyond basic statistics and deliver position-aware, context-sensitive performance ratings using deep learning. It integrates Expected Goals (xG), Expected Assists (xA), and advanced per-90 statistics to generate automated scouting reports through an intuitive web-based interface.

---

🚀 Key Features

• 🎯 Position-Aware Attention Neural Network (PAANN): Custom deep learning model that dynamically prioritizes features based on player position.
• 🌟 xA Prediction: Estimated using a trained Random Forest model based on passing and creative playmaking stats.
• 📊 Advanced Metrics:
  • xG, xA, G–xG, A–xA
  • Per-90 statistics for fair cross-player comparisons
• 📈 Smart Visuals:
  • Radar charts for skill profiles
  • Position heatmaps
  • Automated verdict on performance tier
• 🌐 Clean Web Interface:
  • Built using Flask
  • Interactive scouting report generator

---

🧠 Model Architecture

PAANN – Position-Aware Attention Neural Network

• Position Embedding: Encodes player role into the model
• Feature + Position Concatenation: Adds positional context to raw features
• Attention Mechanism: Weighs features based on role-specific importance
• Encoder: Dense layers with BatchNorm + Dropout for stable learning
• Output: Predicts real-valued player rating (scale of 0–10)

---

📁 Tech Stack

Component	Technology
Model Training	PyTorch
xA Prediction	Random Forest (Sklearn)
Frontend + API	Flask
Visualization	matplotlib, seaborn, mplsoccer
Deployment	Render

---

🛠️ How It Works

1. Enter player performance stats (attacking, passing, defensive)
2. The PAANN model predicts the player's role-aware rating
3. xA is computed using a trained Random Forest model
4. A scouting report is generated with:
   • Rating
   • Radar chart
   • Advanced stats (xG, xA, G–xG, A–xA)
   • Verdict on player performance

---

📊 Model Performance

Metric	Training Set	Validation Set
MSE	0.0040	0.0065
MAE	0.0431	0.0504
R² Score	0.9727	0.9568
Accuracy	94.23%	92.86%
F1 Score	91.52%	85.04%

---

⚠️ Limitations

• No match-by-match or time-series analysis (season-aggregated only)
• Goalkeepers not included
• Approximate xG due to lack of granular shot data
• Single-head attention (multi-head attention in future plans)

---

🔮 Future Enhancements

• Live match feed integration
• Mobile and API version
• Player comparison tools
• Graph Neural Networks and more advanced DL models
• Expanded dataset and real-time scouting

---