Defensive Performance Analysis in European Football using PCA and Multiple Linear Regression

Exploring Defensive Team Profiles of Top 5 European Leagues — 2023/24 Season

A statistical analysis of team-level defensive data from the five major European football leagues (Serie A, Bundesliga, Ligue 1, La Liga, Premier League) in the 2023/24 season, applying Principal Component Analysis (PCA) and Multiple Linear Regression.

---

⚠️ Reproducibility notice

This project was developed in 2024. Since then, FBref has restricted access to several statistical tables that were previously available for free scraping. As a result, the data-loading step in report.Rmd — which relies on worldfootballR::fb_season_team_stats(..., "defense") — no longer works and will return an error or empty data.

Unfortunately the original dataset was not exported at the time and cannot be redistributed. The repository is therefore preserved as a methodological reference: the full analysis pipeline (descriptive statistics, PCA, multiple regression, and diagnostic tests) remains valid and can be adapted to any equivalent dataset obtained through alternative means (e.g. a paid FBref subscription, or a manually exported CSV).

---

📋 Overview

This project explores how and how much top European clubs defend, going beyond simple tallies to identify:

• Tactical profiles of individual teams through descriptive statistics and radar charts.
• Latent defensive dimensions via PCA (e.g., overall defensive intensity, pressing orientation, midfield efficiency, error propensity).
• Linear drivers of tackle success via multiple regression with backward variable selection.

---

🖼️ Visuals

PCA Biplot — UCL vs Non-UCL Teams

Most Champions League qualifiers cluster in the positive PC1–PC2 region, suggesting that top clubs tend to be more active defensively — pressing higher and winning more tackles across all zones of the pitch.

Radar Chart Example

Each team's defensive profile expressed as percentile ranks across all 14 variables.

---

📂 Repository Structure

DefensivePerformance_PCA/
├── report.Rmd          # Main analysis document (R Markdown)
├── report.pdf          # Compiled PDF report
├── R/
│   └── helpers.R       # Reusable utility functions
├── data/
│   └── README.md       # Notes on data provenance and caching
├── output/
│   └── figures/
│       ├── pca_biplot_ucl.png     # PCA biplot — UCL vs non-UCL teams
│       └── radar_example.png      # Radar chart example
└── docs/
    └── variable_glossary.md   # Full variable descriptions

---

📦 Dependencies

All analysis is done in R. Install required packages with:

install.packages(c(
  "tidyverse", "corrplot", "factoextra", "worldfootballR",
  "ggplot2", "GGally", "car", "lmtest", "PerformanceAnalytics",
  "gridExtra", "fmsb", "psych", "scales"
))

Note: worldfootballR scrapes data from fbref.com. An active internet connection is required to reproduce the data-loading step. To avoid repeated scraping, the cleaned dataset can be cached locally — see data/README.md.

---

🚀 Reproducing the Analysis

1. Clone the repository:

   git clone https://github.com/marinoalfonso/DefensivePerformance_PCA.git
   cd DefensivePerformance_PCA

2. Open report.Rmd in RStudio (or any R environment).

3. Install dependencies (see above).

4. Knit the document to PDF or HTML:

   rmarkdown::render("report.Rmd", output_format = "pdf_document")
   # or
   rmarkdown::render("report.Rmd", output_format = "html_document")

---

🔑 Key Findings

Stage	Main Result
Descriptive	Tottenham leads in attacking-third tackles; Juventus leads in dribble-countering %; Liverpool concedes most dribbles (Gegenpressing effect).
PCA	4 PCs explain >75% variance. PC1 = overall defensive intensity; PC2 = pressing height; PC3 = midfield efficiency; PC4 = error propensity.
Regression	Best model for TklWin: tackles in all three field zones (positive) + shots blocked (negative). Adjusted R² > 0.90 after backward selection.

---

📊 Data Source

Data extracted via the worldfootballR R package from fbref.com, a widely used football statistics aggregator. The dataset covers 96 teams (20 per league, 18 for Bundesliga and Ligue 1) with 14 numeric defensive variables per team.

---

👤 Author

Alfonso Marino
GitHub · Feel free to open an issue or submit a PR.

---

📄 License

This project is licensed under the MIT License - see the LICENSE file for details.