Eredivisie player valuation models through the application of machine learning methodologies

This thesis addresses limitations in existing football player valuation models by developing more comprehensive and accurate position-based models using advanced machine learning techniques. Focusing on the Eredivisie, the study integrates features from various categories such as: player characteristics, performance, crowd-judgment, potential and team factors. Conducted in two phases, the research first analyzed the influence of different features on player valuations, followed by testing various machine learning models, including Linear Regression, Ridge, Lasso, PSO-SVR, LightGBM, XGBoost, CatBoost and a Meta-model ensemble. The findings show that models incorporating diverse feature subsets outperform those relying on single sets. PSO-SVR performed best for attackers and midfielders, CatBoost with Bayesian Optimization for defenders and XGBoost for goalkeepers. The study concludes that position-based models leveraging the latest machine learning techniques and use features from all categories significantly enhance predictive accuracy and outperform traditional models. These models provide football clubs with actionable insights for transfer decisions, strategic planning and financial sustainability, ultimately influencing long-term success.