Preventive Models for Respiratory Illness and Common Injuries in Professional Soccer: Learning Algorithms
This study quantified intrinsic risk factors through ensemble machine learning and CNNs (Convolutional Neural Networks) to yield injury and respiratory illness risk in elite soccer athletes. Through python’s scikit learn machine learning library, the researcher selected the best models, identifying soccer players with a high risk of frequent injury and illness types. Injuries and illness from 1,346 male soccer players from 98 professional European teams (2009–2019) were scraped from www.transfermarkt.us, with a total of 123,775 time-loss-days due to injury and 4,556 time-loss-days due to illness recorded. The final model was built with ensemble machine learning techniques, like SMOTE, with a GradientBoostingClassifier as the base classifier. It reported the best evaluation criteria (highest accuracy score: ≈0.80, highest cross-validation score: 0.79 (+/- 0.01)) and the model built with deep learning methods yielded accuracies within the 90-99% range and loss function under 1.0. Hence, the algorithms made proved optimal for prediction and developing WEB/SMS based application: sportinjuri. With moderate to high accuracy scores, the models and derived applications may inform the decisions of coaches, physical trainers and medical practitioners when faced with a player’s illness or injury.