Can we predict archery performance using shooting heart rate and other biophysical parameters? A machine learning approach

Objective: We aimed to develop an ML model to predict the archery performance score using key performance parameters, including the HR values during key phases of shooting, and determine the predictive importance of these parameters to provide personalised training to an archer.

Methods: A total of 32 archers (15 elite and 17 non-elite) shot two sessions of 30 arrows each indoors, wearing an HR chest monitor and were videographed. When each arrow was shot, 11 HR values were identified at different shooting phases. Biophysical parameters consisting of 35 linear variables and second-degree polynomial HR values were used to build ML models in Python. The total scores of Sessions 1 and 2 were used to train and test, respectively. Model performance was evaluated using root mean squared error and highest accuracy. Shapley's additive explanation method was used to show the predictive importance of each variable.

Results: The Cat Boost ML model with the lowest error and better accuracy was used to predict the archery score. Sports age, resting systolic blood pressure, previous competition score, right-hand grip strength, age, HR rate before 2 s of arrow release, waist-to-hip ratio, concentration disruption, trait anxiety and HR after 5 s of release are the top parameters that predicted the score.

Conclusions: This study highlights that the ML approach can be a useful tool to predict top parameters to optimise archery performance. The ML model has been trained and tested for indoor archery settings in a controlled environment, which may be a limitation in generalising this ML model for prediction during outdoor competitive settings.