Applying neuro-fuzzy modeling to evaluate and enhance badminton footwork training

Abstract

Purpose: To design a coach-interpretable movement quality score for badminton footwork training using an adaptive neuro-fuzzy inference system. Methodology: We derived compact biomechanical descriptors of timing, kinematics, dynamics, and stability from benchmark motion recordings and trained a first order Takagi Sugeno ANFIS with Gaussian membership functions and hybrid learning using a 70 15 15 train validation test split with early stopping. Results: The model achieved RMSE 0.074, MAE 0.058, and R2 0.91 on the test set and outperformed linear regression, multilayer perceptron, and support vector regression while preserving transparent fuzzy rules. Conclusion: ANFIS provides accurate and explainable quality estimation. Recommendations: Future work should validate on in-court badminton datasets, expand features for foot contact and center of mass, and deploy lightweight wearable inference for practical real-time feedback.