Research Progress of Sports Injury Prediction Model Based on Machine Learning

This systematic review aims to analyze sports injury prediction models based on machine learning technology in order to provide references for the development of sports injury prediction models. The study employed a literature review method to search the Web of Science. The models consider a range of features, including anthropometry, sports quality, training load, injury history, exercise duration, sleep, genetic information, and academic performance. Feature selection strategies applied include forward selection, Gini coefficient decline, and lasso analysis. The training algorithms utilized are support vector machines, random forest, decision trees, neural networks, and logistic regression. Model performance evaluation methods consist of HoldOut crossover and k-crossover methods, with evaluation metrics varying from AUC, sensitivity, specificity, F1 score to accuracy. Improvements are suggested to include the competition scene and sports recovery indicators in the feature extraction process. Moreover, greater diversity in training sets, utilization of unsupervised and semi-supervised learning algorithms, and implementation of multiple model optimization methods are recommended to enhance the models' robustness and overall performance.