Michal Zareba
Current Organizations / Institutes
The organizations/institutes from the scholar's latest article
Research Subjects
Michal Zareba has a total of 9 co-authors. Here are the top co-authors by number of shared publications:
- Tomasz Piłka
Faculty of Mathematics and Computer Science, Adam Mickiewicz University, 61-614 Poznań, Poland
3 shared publications - Aleksandra Sadurska
Adam Mickiewicz University, Poland
1 shared publication Bartłomiej Grzelak
Adam Mickiewicz University in Poznan
1 shared publication- Bartłomiej Grzelak
Faculty of Mathematics and Computer Science, Adam Mickiewicz University, 61-614 Poznań, Poland
1 shared publication Krzysztof Dyczkowski
Adam Mickiewicz University in Poznan
1 shared publication- Krzysztof Dyczkowski
Adam Mickiewicz University, Poland
1 shared publication
Latest Publications
- Open Access
- 17 November 2025
- Published by International Conference on Information Systems Development
Gradient or Not? Predicting Football Action Sequences Using Boosting vs Neural Networks
This research compares gradient boosting methods and neural network architectures for predicting football action sequences. Using detailed event annotations and spatial-temporal positional data, we evaluate the models’ ability to forecast goal-scoring opportunities several actions in advance. Through feature engineering and ensemble strategies, our results reveal key contextual and spatial factors that influence goal probabilities. Ensemble models combining CatBoost, LightGBM, and XGBoost outperform individual models, achieving an F1 Score of 0.707 and PR AUC of 0.734. These findings can provide valuable insights for real-time match analysis and player evaluation.
- 17 November 2025
- Published by International Conference on Information Systems Development
Open AccessGradient or Not? Predicting Football Action Sequences Using Boosting vs Neural Networks
This research compares gradient boosting methods and neural network architectures for predicting football action sequences. Using detailed event annotations and spatial-temporal positional data, we evaluate the models’ ability to forecast goal-scoring opportunities several actions in advance. Through feature engineering and ensemble strategies, our results reveal key contextual and spatial factors that influence goal probabilities. Ensemble models combining CatBoost, LightGBM, and XGBoost outperform individual models, achieving an F1 Score of 0.707 and PR AUC of 0.734. These findings can provide valuable insights for real-time match analysis and player evaluation.
- Open Access
- 17 November 2025
- Published by International Conference on Information Systems Development
Leveraging GPS Data and Attention-based BiLSTM for Injury Prediction in Professional Football
Injuries pose a significant challenge in professional football, affecting player availability, team performance, and club finances. Accurate prediction of injury risk is crucial for implementing effective prevention strategies. This study develops a deep learning model to predict the likelihood of injury in professional football players using data collected through Catapult Sports tracking devices. The research is carried out in collaboration with KKS Lech Poznań, a Polish professional football club. The proposed model architecture combines bidirectional long- and short-term memory (BiLSTM) networks with an attention mechanism to learn from the time series data and predict player injury risk. The model is trained on sequences of Catapult data spanning 14 days before each recorded injury or non-injury event. To address class imbalance, a custom loss function was implemented that balances focal loss and the F_beta score. The model's performance is evaluated on an independent test set, achieving a specificity of 0.90, an accuracy of 0.90, and a recall of 0.40.
- 17 November 2025
- Published by International Conference on Information Systems Development
Open AccessLeveraging GPS Data and Attention-based BiLSTM for Injury Prediction in Professional Football
Injuries pose a significant challenge in professional football, affecting player availability, team performance, and club finances. Accurate prediction of injury risk is crucial for implementing effective prevention strategies. This study develops a deep learning model to predict the likelihood of injury in professional football players using data collected through Catapult Sports tracking devices. The research is carried out in collaboration with KKS Lech Poznań, a Polish professional football club. The proposed model architecture combines bidirectional long- and short-term memory (BiLSTM) networks with an attention mechanism to learn from the time series data and predict player injury risk. The model is trained on sequences of Catapult data spanning 14 days before each recorded injury or non-injury event. To address class imbalance, a custom loss function was implemented that balances focal loss and the F_beta score. The model's performance is evaluated on an independent test set, achieving a specificity of 0.90, an accuracy of 0.90, and a recall of 0.40.
- Open Access
- 21 August 2025
- Published by Springer Nature in Research Square
Decoding Defensive Performance: A MachineLearning Approach to Football Player Valuation
Analyzing defensive actions, which have traditionally received less attention thanoffensive metrics, is a significant challenge in football analytics. This researchpresents an innovative methodology that utilizes XGBoost and deep neuralnetworks to evaluate defensive performance using metrics such as On-Ball Value(OBV), Valuing Actions by Estimating Probabilities (VAEP), and eXpectedThreat (xT). The study proposes a machine learning-based framework forevaluating defensive player value. A case study using expert ratings and marketvalues from the Polish PKO BP Ekstraklasa demonstrates the method’seffectiveness. The results advance the field of sports analytics by addressing thepersistent problem of accurately valuing the defensive contributions of footballplayers.
- 21 August 2025
- Published by Springer Nature in Research Square
Open AccessDecoding Defensive Performance: A MachineLearning Approach to Football Player Valuation
Analyzing defensive actions, which have traditionally received less attention thanoffensive metrics, is a significant challenge in football analytics. This researchpresents an innovative methodology that utilizes XGBoost and deep neuralnetworks to evaluate defensive performance using metrics such as On-Ball Value(OBV), Valuing Actions by Estimating Probabilities (VAEP), and eXpectedThreat (xT). The study proposes a machine learning-based framework forevaluating defensive player value. A case study using expert ratings and marketvalues from the Polish PKO BP Ekstraklasa demonstrates the method’seffectiveness. The results advance the field of sports analytics by addressing thepersistent problem of accurately valuing the defensive contributions of footballplayers.
- Open Access
- 9 September 2024
- Published by International Conference on Information Systems Development
Improving The Evaluation of Defensive Player Values with Advanced Machine Learning Techniques
Quantifying defensive actions, which offensive indicators have historically overshadowed, is challenging in football analysis. This study presents a novel approach using XGBoost and neural networks to evaluate defensive play using On-Ball Value (OBV), Valuing Actions by Estimating Probabilities (VAEP), and eXpected Threat (xT) indicators. The proposed evaluation of Defensive Player Value using machine learning techniques is presented. A comparative assessment of expert ratings and market values in a Polish PKO BP Ekstraklasa case study highlights the method's effectiveness. The research contributes to the development of sports analytics by addressing the long-term challenge of evaluating the defensive play of football players.
- 9 September 2024
- Published by International Conference on Information Systems Development
Open AccessImproving The Evaluation of Defensive Player Values with Advanced Machine Learning Techniques
Quantifying defensive actions, which offensive indicators have historically overshadowed, is challenging in football analysis. This study presents a novel approach using XGBoost and neural networks to evaluate defensive play using On-Ball Value (OBV), Valuing Actions by Estimating Probabilities (VAEP), and eXpected Threat (xT) indicators. The proposed evaluation of Defensive Player Value using machine learning techniques is presented. A comparative assessment of expert ratings and market values in a Polish PKO BP Ekstraklasa case study highlights the method's effectiveness. The research contributes to the development of sports analytics by addressing the long-term challenge of evaluating the defensive play of football players.