Injury prediction model for lower-limb sports injuries: A novel machine learning-based approach

Area of Science:

• Sports Medicine
• Biomechanical Engineering
• Data Science in Sports

Background:

• Sports injuries result from complex interactions of risk factors.
• Identifying at-risk athletes is crucial for effective injury prevention.
• Machine learning and AI offer advanced tools for injury prediction.

Purpose of the Study:

• To develop an effective injury prediction model for lower-limb sports injuries.
• To compare the efficiency of different machine learning models for injury prediction.

Main Methods:

• Collected data from 120 male university athletes across football, cricket, and basketball.
• Recorded 44 variables including mental and physical screening, and previous season injury data.
• Built and compared various machine learning models for injury prediction.

Main Results:

• Ankle injuries were most common (52.6%), followed by hamstring (23.6%), foot (13.3%), and knee (10.5%).
• The K Nearest Neighbour (KNN) algorithm with Random Oversampling (ROS) achieved the highest accuracy (AUC=0.87, TPR=100%, TNR=62.5%).
• Thirty-two injuries were used to construct the final predictive model.

Conclusions:

• The KNN model with ROS demonstrates strong predictive accuracy for athlete injuries in football, cricket, and basketball.
• This model can be a valuable tool for enhancing sports injury prevention strategies.
• Further prospective studies are recommended to validate the model's efficacy.