Match Outcome Prediction Based on Fitness and Motor Ability Parameters in Youth Badminton

In this chapter, we investigated the prediction of match outcomes (winner vs. loser) among youth badminton players using physical fitness and motor ability indicators. Sixty-seven athletes were assessed on a set of 13 indicators, including balance, coordination, muscular endurance, flexibility, power, agility, reaction time, and grip strength. Total points gathered during the badminton competition were used to categorise the winners and losers. Due to class imbalance (63% winners, 37% losers), the synthetic minority oversampling technique (SMOTE) was applied during training to improve minority-class learning. Several machine learning models were evaluated, including decision tree, k-nearest neighbours (kNN), random forest, and a stacking ensemble. The findings demonstrated that conventional classifiers, even after tuning and SMOTE balancing, achieved moderate performance (random forest accuracy = 0.50; kNN accuracy = 0.43), while decision tree exhibited poor generalisation. Conversely, the stacking ensemble substantially outperformed all individual models, achieving 0.79 accuracy, 0.83 balanced accuracy, and 0.78 macro F1-score, with perfect recall for the minority class. These findings suggest that ensemble blending strategies effectively address class imbalance and capture complex, nonlinear patterns in performance data, offering a promising approach for match outcome prediction in youth badminton.