Predicting Soccer Transfer Values
In club soccer, acquiring players works differently than in American sports. In baseball, basketball, and football players are acquired as free agents or traded for other players. In soccer, most players change teams by being bought and sold. This process is referred to as the transfer market. Properly assessing the value of your own players as well as your targets is a massive competitive advantage for clubs. In recent years, Liverpool has been lauded for its astute activity in the transfer market. They were rewarded in the summer of 2020 with their first-ever Premier League Championship.
I decided to build a machine learning model that predicted the transfer values of players based on their statistics. In the end, I had mixed results. My model was quite accurate when it predicted that a player would be very expensive, however, it often significantly underestimated the price of defenders.
I started by collecting my data. I gathered transfer data from transfermarkt.com. I scraped player statistics from fbref.com. Because the statistics for goalkeepers are completely different from outfield players I removed all goalies from my dataset. As a next step, I plan on creating a model that will predict transfer values for goalies. I combined my transfer data with player statistics and then started creating models.
My data was heavily positively skewed. While it had mean and median values of 8.2 and 4.1 million pounds respectively the largest transfer was 130.5 million pounds.
I ran a dummy model that always predicted the mean transfer value. I then ran linear regression, ridge and lasso regressions, support vector machines, decision trees, random forest, and xgboost models. I made sure I was using the best possible hyperparameters with grid search. My best model turned out to be the random forest model.
