Football Transfers Network Analisys

This Repository refers to the Final Project of the course Data Driven Models for Complex Systems (DDMCS) at University Sapienza of Rome 2022/2023.

Abstract

The project examined the European football market network of the top 7 leagues over the years. The process involved data extraction from Transfer Markt by web scraping, followed by graph analysis and temporal comparison.

Files:

• Code_R.R contains all the code used to carry out all the various analyses performed in this work;
• Functions.R contains some functions defined here that are used in the Code_R.R file ;
• PreProcessing.ipynb notebook contains all the code used in preprocessing the data;
• Scarping.ipynb notebook contains all the code to scrape the extra data of transfer markt;
• datasets_cleaned folder has all the datasets used in this work;
• Imgs folder includes some of the Image and graphs carried out in my work.

Dataset

The data comes from Kaggle , scraped from Transfer Markt. The datasets regards all the transfer operation in the major 7 European leagues from the 1992/1993 season to the 2021/2022 season. The raw data were extremely dirty so a massive and detailed cleaning operation was necessary.

Cleaning procedures

1. Scrape data from Transfer Markt to build two types of dictionaries:

• one that maps every team name for every season its long name to the short one. (ex: "Juventus FC" : "Juventus").
• one that maps each club not in the 7 top leagues in its continent/country. (ex: "Santos" : "South America").

2. Delete all the "out" and "End of loan" operations to avoid duplicate transactions.

3. Check and correct some typos in original data.

Graph Builindg

Every team is a Node. If team A buys a player from team B, they are connected with a link, weighted with the fee and directed to B. For every season:

• Nodes: 134
• Links: a) ~900 if one considers all transactions b) ~600 if one considers only transactions from team of Top 7 European Leagues

Results

Degree

The degree of a team, represents how many different teams it has had negotiations with (both in and out). The Average degree of the teams has more or less doubled over the seasons (from 4 to 7.5). The maximum degree of clubs has increased over time, although the effects are less noticeable than the average degree. After all, a team cannot make too many different deals per season, as 11 players always take the field.

In the figure above we can notice:

• Most of the operations are loans
• There are only few Transfers with a fee > 5 Milion
• Each team on average is involved in deals with 4 other clubs.
• Only a few clubs do business with more than 10 teams

In the recent years a lot of things are changed:

• There are some edges with a high charge
• Three purchases exceed 100 million
• Each team on average is involved in deals with 9 other clubs.
• 43 clubs over 134 do business with more than 10 teams.
• Still a lot of transactions are loans

Until 2000, no purchase had ever exceeded 50 million. After the purchase of G. Bale, the Market exploded: at least 75 million was spent each year on a deal. Neymar to PSG for 225 million is the most expensive transfer in football history.

Measures on the overall network

• The Diameter has decreased quite a bit over the seasons (from 20 to 10 and in 2017/2018 was equal to 8).
• Also the Mean distance decreased a lot.
• The Density has increased, but it's still very low.
• The Mean clustering coefficient has varied over the seasons, but on average, apart from the last 3 years, it has not changed much from its initial value.

Measures averaged by each league sub-network

• All four measures, calculated inside the leagues remain very constant over the years.
• We can see that the Diameter and the Mean Distance are lower than those calculated in the entire network.
• The Density and The Mean clustering coefficient are higher than those in the whole network.

Modularity and Communities

• In the initial four seasons, the score was very high.
• The results of all three cases calculated are in agreement with each other, and show a continuous decrease in this value.
• The market has become more and more global.

• Very closed structure.
• Few clubs buy outside their league
• Transfers are fewer and the amounts are very low.

• The structure is very open, it almost looks like one league.
• There are many transfers and prices are often high.

Centrality

It is interesting to note that, although each season has different negotiations, the nodes with the highest betweness centrality are very often the same. In addition, almost all of them are mid-to-high-end clubs: this makes his players attractive to all the big teams in Europe but also to the lower-end ones.

League Comparision

For the first three seasons more than 50% of the money spent in European football came from Italian teams, from 2005 onwards this amount dropped below 25 per cent. Since then its place has been taken by the Premier league. In recent years, more than half of the money spent in Ligue 1 comes from PSG. Covid-19 stopped the insane growth of spending in all leagues except the Premier league.

Conlcusion

The market has become increasingly global over the years. This is evidenced by the decreasing diameter, average distance and modularity. Over the course of the seasons, the changing balance of power between the leagues can also be seen in the money spent by the various clubs in them. Furthermore, since 2013 there has been a strong price increase. Some clubs like Roma, Porto and Chelsea have, in many seasons, always extremely high betweness centrality values.