BreakLines: Uncovering Tactical Line-Breaking Passes

This repo consists of codes, presentations and other details of the Breaklines Project

An unsupervised, interpretable framework for detecting and analysing line-breaking passes (LBPs) using spatiotemporal football data.

Overview

Line-breaking passes (LBPs) are a cornerstone of vertical progression and attacking structure in football. This project introduces a clustering-based method to detect LBPs from synchronised event and tracking data, modelling opponent defensive shape dynamically at the moment of each pass.

BreakLines further introduces two key tactical metrics:

• SBR (Space Buildup Ratio): Quantifies the spatial advantage gained from a pass.
• LBPCh¹ & LBPCh²: Capture direct and chained progression sequences that culminate in goal-scoring attempts.

Our approach combines spatial logic with tactical relevance to uncover the structure behind dangerous build-up sequences — without relying on proprietary labels or pre-defined formations.

📄 Paper

Read the full paper here:
➡️ Through the Gaps: Uncovering Tactical Line-Breaking Passes with Clustering

🔧 Features

• 🧠 Unsupervised Clustering: Dynamic vertical segmentation of defensive lines using K-Means with adaptive k (≥2 clusters).
• 📊 SBR Metric: Interpretable spatial impact score based on area expansion around receivers.
• 🔄 Chain Analysis: Detection of chained LBPs (LBPCh²) for sustained vertical build-ups.
• ⚽ Player/Team Analysis: Identify top line-breakers, space creators, and tactical contributors.

🗃️ Dataset

Uses the PFF FC World Cup 2022 Dataset which includes:

• Event data (passes, duels, shots)
• Tracking data at 29.97Hz for all players and ball
• Roster, meta, and alignment information for full context

📂 Folder Structure

breaklines/
│
├── figures/             # Paper and visualization assets
├── codes/               # Jupyter analysis notebooks
├── src/                 # Core scripts for clustering, SBR, LBPCh
├── videos/              # Video recording for the presentations
└── README.md

📌 Citation

@article{karakus2025breaklines,
  title={Through the Gaps: Uncovering Tactical Line-Breaking Passes with Clustering},
  author={Karakuş, Oktay and Arkadaş, Hasan},
  journal={arXiv preprint arXiv:2506.06666},
  year={2025}
}

Made with ❤️ by Dead Ball Analytics and Cardiff University

Contact: Dr Oktay Karakus: [email protected]