Enhanced Maturity Status Classification Using SMOTE-Augmented Machine Learning and Kolmogorov–Arnold Networks for Youth Fitness and Neuromuscular Performance

In this chapter, we examined the feasibility of predicting biological maturity status (early, average, and late) in youth badminton players using neuromotor control and physical fitness indicators. A total of sixty-seven youth badminton players were assessed on several fitness and neuromuscular control-related variables as independent variables. Peak height velocity was used to determine the maturity status of the players, and k-means clustering was used to group the players into different maturity levels. Upon the clustering analysis, severe class imbalance was detected, viz. late: 64%, average: 24%, and early: 12%. To address the inherent class imbalance, the synthetic minority oversampling technique (SMOTE) was applied during model training. Among the tested algorithms, the Kolmogorov–Arnold networks (KAN) combined with SMOTE achieved the highest performance (accuracy = 0.71; macro F1 = 0.47), outperforming random forest in overall classification and particularly improving recognition of average maturity. Both models struggled to identify early maturity, reflecting limited sample representation and overlapping class characteristics. These findings suggest that neuromotor and fitness attributes provide meaningful signals for maturity estimation and highlight the potential of advanced nonlinear models with oversampling strategies to enhance predictive accuracy in youth athlete development research.