Predicting Football Player Transfer Values Using Bagging and Hybrid Machine Learning Approaches

Abstract

Accurately assessing a football player's market value is essential for enabling informed decision-making by clubs, agents, and investors during player transfers, contract negotiations, and strategic investment planning. In this context, machine learning (ML) algorithms offer a robust framework for analyzing historical data, performance indicators, and market dynamics to produce realistic valuations. These datadriven methods assist in identifying undervalued opportunities and flagging overpriced players, thereby enhancing the overall efficiency of transfer market operations. The dataset employed in this research includes a comprehensive set of player-related features such as age, weight, weak foot rating, preferred foot, and international reputation, among others. These attributes collectively contribute to a detailed profile of each player's capabilities and market relevance. The objective of this study is to develop reliable and accurate predictive models that estimate player market values by leveraging advanced machine learning techniques, thereby improving upon traditional, subjective valuation approaches. Several regression-based models were explored, including Bagging Decision Tree Regression (Bg_DT), and Bagging Support Vector Regression (Bg_SVR). To further enhance model performance, optimization algorithms such as Motion-encoded Particle Swarm Optimization (Motion-encoded PSO) and the Red Deer Algorithm (RDA) were applied for hyperparameter tuning. Among the evaluated models, the Bagging Decision Tree optimized with Motion-encoded PSO (Bg_DT- Motion-encoded PSO) demonstrated superior performance. It achieved the lowest Root Mean Squared Error (RMSE) and the highest coefficient of determination (R²) across both validation and testing phases. Specifically, the Bg_DT- Motion-encoded PSO model yielded an RMSE of 533×10⁵ and an R² of 0.962 during validation, indicating strong predictive accuracy and generalization capability. These findings underscore the effectiveness of ensemble learning techniques—particularly Bagging Decision Trees—in conjunction with advanced metaheuristic optimizers like Motion-encoded PSO, for accurately estimating football player market values.