Scholarlink

<p>In football economics, a player’s transfer market value extends beyond performance metrics, with popularity playing a crucial role in clubs’ decisions. Reputation indexes, reflecting a player’s standing in the industry, are derived from various sources. Traditional metrics include goals, assists, and defensive prowess, while social media activity (likes on Facebook and Instagram), press citations, and Wikipedia page views add a new dimension. This study utilized Fédération Internationale de Football Association 19 data and a real-world statistical dataset, encompassing 54 features for 491 players across various leagues. After adding valuable data and removing ineffective features and outliers, two filtering-based feature selection methods identified the 20 most critical features for predicting market value. The study applied Extreme Gradient Boosting and Adaptive Boosting regression models, along with their hybrid forms optimized by metaheuristic algorithms. The Extreme Gradient Boosting optimized with the Ali Baba and Forty Thieves algorithm model showed the best performance, with a 99% match to actual values and a misestimation of around €1.9 million. Ensemble models, averaging predictions from all hybrid models, provided reliable market value estimates. These insights help managers make informed decisions to improve team performance and secure financial benefits for the club.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Comparative Analysis of Hybrid and Ensemble Machine Learning Approaches in Predicting Football Player Transfer Values

  • 期刊论文

  • 发表日期：

    2025年3月31日

查看全文
  • Wenjing Zhang,
  • Dan Cao

摘要

In football economics, a player’s transfer market value extends beyond performance metrics, with popularity playing a crucial role in clubs’ decisions. Reputation indexes, reflecting a player’s standing in the industry, are derived from various sources. Traditional metrics include goals, assists, and defensive prowess, while social media activity (likes on Facebook and Instagram), press citations, and Wikipedia page views add a new dimension. This study utilized Fédération Internationale de Football Association 19 data and a real-world statistical dataset, encompassing 54 features for 491 players across various leagues. After adding valuable data and removing ineffective features and outliers, two filtering-based feature selection methods identified the 20 most critical features for predicting market value. The study applied Extreme Gradient Boosting and Adaptive Boosting regression models, along with their hybrid forms optimized by metaheuristic algorithms. The Extreme Gradient Boosting optimized with the Ali Baba and Forty Thieves algorithm model showed the best performance, with a 99% match to actual values and a misestimation of around €1.9 million. Ensemble models, averaging predictions from all hybrid models, provided reliable market value estimates. These insights help managers make informed decisions to improve team performance and secure financial benefits for the club.