Swing characteristics affecting carry distance in golf using explainable AI

Abstract

Abstract The application of explainable AI is effective for interpreting nonlinear and complex models, such as those governing golf-ball trajectory. This study aimed to identify the swing characteristics that influence carry distance using explainable AI. A machine learning model using Extreme Gradient Boosting (XGB) was constructed from club kinematic data and mass data of 27 golfers collected by a launch monitor. Permutation Feature Importance analysis and SHAP (SHapley Additive exPlanations) interaction value were used to explain the model’s predictions. The coefficient of determination of the XGB test data was 0.71 ± 0.25. The root-mean-square error of the XGB test data was 15.7 ± 6.4. The predictive model showed that club-head speed, club mass, and launch angle were particularly important. Among all feature combinations, the SHAP interaction value for club-head speed and club mass in the XGB model was the highest (3.0). The results indicate that golfers with long carry distances use clubs weighing 0.32 kg or more, achieve a club-head speed of 45 m/s or higher. To increase carry distance, golfers and coaches may need to focus more on increasing club-head speed and reassessing club mass, rather than on the ball’s spin characteristics.