Predicting badminton outcomes through machine learning and technical action frequencies.

The application of machine learning techniques to predict badminton match outcomes through the analysis of technical actions seems to represent an area that has not yet been extensively investigated within the existing body of research. This study aims to interpret this phenomenon by developing a predictive model based on the frequency of technical actions, utilizing machine learning techniques. Focusing on international competitions from 2019 to 2023, we collected data on 23 distinct technical actions (e.g., Net Front, Slice/Drop, Push) to construct predictive models. The study distinguishes itself by employing a Random Forest algorithm to ascertain the significance of each technical action, utilizing forward stepwise selection and 5-fold cross-validation for feature refinement. SHAP value analysis further validated the pivotal roles of 'Net Front', 'Slice/Drop', and 'Push' across both sexes, linking higher frequencies of 'Net Front' with increased match-winning probabilities. Model validation on a test set demonstrated effective performance in both sexes, with the model based on male data exhibiting higher accuracy and predictive values, surpassing the performance of the female data model. This comprehensive examination, grounded in quantitative analysis, not only enhances our understanding of badminton gameplay dynamics but also offers valuable insights for coaching strategies and training methodologies. To extend the applicability of our findings and facilitate user engagement, we developed a web application based on our model. This platform enables players, coaches, and researchers to input player characteristics and receive strategic recommendation through an intuitive interface, further leveraging the machine learning model's capabilities to support tactical decision-making before badminton competitions.