# Prediction of ACL injury incidence and analysis of key features in basketball players based on multi-algorithm models

## TL;DR

This study uses machine learning to predict ACL injuries in male basketball players and identifies key biomechanical factors that increase injury risk.

## Contribution

The study introduces a multi-algorithm approach to predict ACL injuries and identifies critical biomechanical features specific to basketball.

## Key findings

- The random forest model achieved the highest predictive accuracy (AUC-ROC = 0.80) for ACL injury prediction.
- Key biomechanical features included increased knee flexion moment and backward ground reaction force during emergency stops.
- Biomechanical testing based on sport-specific movements improves ACL injury risk prediction.

## Abstract

Basketball players are a high-risk group for anterior cruciate ligament (ACL) injuries. This study aimed to identify the critical factors contributing to ACL injuries in male basketball players and evaluate the performance of machine learning (ML) algorithms in injury prediction.

This study protocol was registered with International Standard Registered Clinical/soCial sTudy Number (ISRCTN) (Registration number: 18009799). A total of 104 male collegiate basketball players volunteered to participate in this study. Data on the athletes’ profile, physical functions, basketball-specific skills, biomechanics, and electromyography (EMG) of seven lower limb muscles during unanticipated side-cutting maneuvers were collected. A 12-month follow-up was conducted to compare these variables between the injured (n = 11) and non-injured (n = 93) groups. Only the variables with significant differences between the groups were included in the predictive modeling.

The performance of machine learning models in predicting ACL injury risk was assessed using the area under the curve (AUC) of the receiver operating characteristic (ROC). The AUC-ROC values ranged from 0.66 to 0.80, with the random forest algorithm achieving the highest performance (AUC-ROC = 0.80). The most influential predicting feature observed during the emergency stop phase, included a greater knee flexion moment, reduced knee flexion angle, increased backward ground reaction force, and increased activation of the vastus lateralis muscle.

The random forest model demonstrated superior predictive performance, providing valuable insights into the key risk factors associated with ACL injury among male basketball players. This study highlighted the importance of biomechanical testing based on sport-specific movements to accurately predict the ACL injury risk.