CEU Electronic Theses and Dissertations, 2025
| Author | Bachvarov, Momchil |
|---|---|
| Title | Methodological Challenges in Rare Event Prediction: A Case Study of Time Series Modeling for Sports Injuries |
| Summary | Injuries are an inevitable part of competitive sports but thanks to modern machine learning approaches and the vast amounts of data available, new opportunities to analyze and potentially predict injury risk have emerged. This thesis investigates the methodological challenges of predicting rare events in time series data through a case study on sport injuries. A dataset collected from a track and field team over a seven-year period, was used to train several machine learning methods – including LSTM, GRU and hybrid CNN+GRU architectures. Rather than focusing solely on performance optimization the study explored the methodological and theoretical challeges of predicting rare events. Each modeling decision such as class imbalance handling, feature representation, and validation strategy is analyzed in term of its impact. The findings highlight the fragility of rare event prediction pipelines and emphasize the importance of methodological choices over the model’s complexity. The goal of the work is to contribute to a better understanding of practical and theoretical limits of machine learning for injjury prediction, while exploring the challenges of rare event prediction in a time series context. |
| Supervisor | Gábor Békés |
| Department | Undergraduate Studies BA |
| Full text | https://www.etd.ceu.edu/2025/bachvarov_momchil.pdf |
Visit the CEU Library.
© 2007-2025, Central European University