Classification of External Load Based on Fitness and Motor Ability Parameters in Youth Badminton

In this study, we investigated the application of machine learning models to classify external load categories in youth badminton athletes based on fitness and motor ability tests. Seventy-three youth players (49 males, 24 females; age: 14.45 ± 1.92 years; badminton experience: 6 ± 2.2 years) participated in the study. External load was quantified using Xsens sensors, capturing tri-axial acceleration during competition, whilst the Louvain clustering was used to group the external load values into meaningful groups, i.e. high and low load players. Several machine learning models were trained to classify the external load grouping of the players. Tree-based ensemble models, particularly random forest and eXtreme gradient boosting, demonstrated superior performance in capturing nonlinear relationships between neuromuscular capacities and external load classification compared to logistic regression, support vector machine and k-nearest neighbour. Feature importance analysis identified balance, flexibility, trunk and leg endurance, and change-of-direction speed as key predictors. Feature selection improved interpretability and, in some cases, enhanced model accuracy, with random forest benefiting most from dimensionality reduction. These findings highlight the potential of ensemble learning combined with targeted predictor selection to optimise athlete monitoring and inform evidence-based training strategies in youth badminton.