11/29/2024
Michael's project supports athletes.
Sports play will result in injury for most athletes. Reinjury is understudied yet is a common pitfall in reducing continued and peak sports performance. This study aimed to 1) develop a machine learning model to accurately predict the probability of reinjury in competitive runners, 2) determine factors most associated with predicting the likelihood of reinjury, and 3) identify the time frame of a recurring injury after an initial injury.
F1 score was calculated to determine the best machine learning model of six algorithms. Feature ranking was completed. A time-to-event analysis was conducted to predict time to reinjury using a Cox proportional hazard (CoxPH) model. Schoenfeld residual and Delta-beta residual analyses were performed to determine the model assumption. Logistic regression showed the highest performance.
The most important feature for determining reinjury risk is max recovery. Max recovery (p=0.01), slope max km Z5-T1-T2 one day (p=0.02), and slope number of tough sessions (p=0.03) were determined to have a statistically significant impact on determining time to reinjury. This study can help identify high-risk athletes for reinjury, allowing physical therapists and athletic trainers to take disciplinary action against reinjury.
https://www.iresearchcorporation.com