Sports injury prediction model based on machine learning
- VernacularTitle:基于机器学习的运动损伤预警模型
- Author:
Mengli WEI
1
,
2
;
Yaping ZHONG
;
Huixian GUI
;
Yiwen ZHOU
;
Yeming GUAN
;
Shaohua YU
Author Information
- Keywords: sports injury; injury warning; injury prevention; intelligent warning; machine learning; deep learning; artificial intelligence; sports
- From: Chinese Journal of Tissue Engineering Research 2025;29(2):409-418
- CountryChina
- Language:Chinese
- Abstract: BACKGROUND:The sports medicine community has widely called for the use of machine learning technology to efficiently process the huge and complicated sports data resources,and construct intelligent sports injury prediction models,enabling accurate early warning of sports injuries.It is of great significance to comprehensively summarize and review such research results so as to grasp the direction of early warning model improvement and to guide the construction of sports injury prediction models in China. OBJECTIVE:To systematically review and analyze relevant research on sports injury prediction models based on machine learning technology,thereby providing references for the development of sports injury prediction models in China. METHODS:Literature search was conducted on CNKI,Web of Science and EBSCO databases,which mainly searched for literature related to machine learning techniques and sports injuries.Finally,61 articles related to sports injury prediction models were included for analysis. RESULTS AND CONCLUSION:(1)In terms of external risk feature indicators,there is a lack of competition scenario indicators,and the inclusion of related feature indicators needs to be further improved to further enrich the dimensions of the dataset for model training.In addition,the inclusion feature weighting methods of the sports injury prediction model are mainly based on filtering methods and the use of embedding and wrapping weighting methods needs to be strengthened in order to enhance the analysis of the interaction effects of multiple risk factors.(2)In terms of model body training,supervised learning algorithms become the mainstream choice.Such algorithms have higher requirements for the completeness of sample labeling information,and the application scenarios are easily limited.Therefore,the application of unsupervised and semi-supervised algorithms can be increased in the later stage.(3)In terms of model performance evaluation and optimization,the current studies mainly adopt two verification methods:HoldOut crossover and k-crossover.The range of AUC values is(0.76±0.12),the range of sensitivity is(75.92±11.03)%,the range of specificity is(0.03±4.54)%,the range of F1 score is(80.60±10.63)%,the range of accuracy is(69.96±13.10)%,and the range of precision is(70±14.71)%.Data augmentation and feature optimization are the most common model optimization operations.The accuracy and precision of the current sports injury prediction model are about 70%,and the early warning effect is good.However,the model optimization operation is relatively single,and data augmentation methods are often used to improve model performance.Further adjustments to the model algorithm and hyperparameters are needed to further improve model performance.(4)In terms of model feature extraction,most of the internal risk profile indicators included are mainly based on anthropometrics,training load,years of training,and injury history,but there is a lack of sports recovery and physical function indicators.
