Scholarlink

Background <p>Predicting team performance in football has become an increasingly important topic in sports analytics and performance analysis, particularly for club management, technical planning, and strategic decision-making. This study aims to predict end-of-season team points using Serie A data from the 2015–2025 period and to generate forward-looking performance forecasts for the 2026 season.</p> Methods <p>The study combines financial, sporting, and managerial variables within a Bayesian machine learning framework and compares different statistical and machine learning methods for football performance prediction. Nine independent variables were included in the analysis: squad market value, previous season points, net transfer expenditure, number of new transfers, managerial change, proportion of foreign players, average age, Serie A experience, and participation in UEFA competitions. Variable importance was evaluated using the Bayesian Model Averaging (BMA) approach. Linear Regression, Elastic Net, Random Forest, Gradient Boosting, and Bayesian Additive Regression Trees (BART) models were applied and compared using MAE, RMSE, and R<sup>2</sup> performance measures.</p> Results <p>The BMA results showed that squad market value, managerial change, and participation in UEFA competitions were the most important variables associated with team performance. Among all prediction models, the BART model achieved the best predictive performance, with the lowest error level (RMSE = 7.28) and the highest explanatory power (R<sup>2</sup> = 0.863). In addition, an ensemble forecasting approach was used to provide a more balanced interpretation of model uncertainty. Forecasts for 2026 suggest that teams with high squad value and technical stability, particularly Inter, are expected to occupy the top positions in the league standings.</p> Conclusions <p>The findings indicate that financial structure, managerial stability, and participation in European competitions play an important role in football performance. Overall, the study demonstrates that Bayesian and machine learning–based approaches can provide effective results in football performance prediction. The proposed framework may contribute to the literature on football performance analysis and provide useful insights for strategic planning and performance management in professional football clubs.</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

A Bayesian machine learning framework for forecasting serie a team performance

  • 期刊论文

  • 出版模式：

    开放获取

  • 发表日期：

    2026年7月28日

查看全文
  • Tuba Koç,
  • Mehmet Ali Cengiz,
  • Haydar Koç,
  • Furkan Koçak

摘要

Background

Predicting team performance in football has become an increasingly important topic in sports analytics and performance analysis, particularly for club management, technical planning, and strategic decision-making. This study aims to predict end-of-season team points using Serie A data from the 2015–2025 period and to generate forward-looking performance forecasts for the 2026 season.

Methods

The study combines financial, sporting, and managerial variables within a Bayesian machine learning framework and compares different statistical and machine learning methods for football performance prediction. Nine independent variables were included in the analysis: squad market value, previous season points, net transfer expenditure, number of new transfers, managerial change, proportion of foreign players, average age, Serie A experience, and participation in UEFA competitions. Variable importance was evaluated using the Bayesian Model Averaging (BMA) approach. Linear Regression, Elastic Net, Random Forest, Gradient Boosting, and Bayesian Additive Regression Trees (BART) models were applied and compared using MAE, RMSE, and R2 performance measures.

Results

The BMA results showed that squad market value, managerial change, and participation in UEFA competitions were the most important variables associated with team performance. Among all prediction models, the BART model achieved the best predictive performance, with the lowest error level (RMSE = 7.28) and the highest explanatory power (R2 = 0.863). In addition, an ensemble forecasting approach was used to provide a more balanced interpretation of model uncertainty. Forecasts for 2026 suggest that teams with high squad value and technical stability, particularly Inter, are expected to occupy the top positions in the league standings.

Conclusions

The findings indicate that financial structure, managerial stability, and participation in European competitions play an important role in football performance. Overall, the study demonstrates that Bayesian and machine learning–based approaches can provide effective results in football performance prediction. The proposed framework may contribute to the literature on football performance analysis and provide useful insights for strategic planning and performance management in professional football clubs.