캣부스트 기반에서 축구 선수의 장기부상 예측모델: 예후예측 모델과 조기예측 모델의 비교분석

초록

Recently, research on predicting sports injuries using machine learning has been actively conducted, but most of them focus on predicting the severity of long-term injuries or comparing differences in predictive models according to the timing of information acquisition. The purpose of this study is to develop a CatBoost-based long-term injury prediction model using injury data of European professional soccer players, and to compare and analyze the performance of a prognostic prediction model using post-injury information and an early prediction model that can be used immediately after injury. For the study, 15,603 European professional football player injury data collected from 2020 to 2025 were used. Long-term injuries were defined as cases where the number of days of absence was more than 28 days, and two predictive models were constructed using the CatBoost algorithm. Model performance was evaluated using Accuracy, Precision, Recall, F1-score, and ROC-AUC, and the influence of major predictors was analyzed through SHAP (Shapley Additive Explanations) analysis. As a result of the study, the prognostic prediction model showed very good predictive performance as it recorded Accuracy 90.0%, F1-score 86.4%, and ROC-AUC 96.4%. The early prediction model was shown to be Accuracy 68.4%, F1-score 64.0%, and ROC-AUC 76.8%, and the performance was slightly lower than that of the prognostic prediction model, but the possibility of it as a predictive model that can be used immediately after the actual injury occurred was confirmed. In addition, in the results of variable importance and SHAP analysis, Games missed and Injury were found to be key variables that determine long-term injury, and in the early prediction model, the Injury variable was identified as the most important explanatory variable. This study is differentiated from previous studies in that it simultaneously compared and analyzed prognosis and early prediction models for predicting long-term injuries. These results can be used to establish rehabilitation plans for sports medical staff and coaching staff, to determine when to return to players, and to support data-based decision-making.

키워드