Treffer: The Analysis of Basketball Endurance Training and Fatigue Assessment Based on Nonlinear Neural Network Model.

This study develops a nonlinear neural network-based system for assessing fatigue states, aiming to support personalized endurance training and real-time health monitoring for basketball athletes. By collecting multidimensional physiological data specific to basketball players − including heart rate, oxygen consumption, and movement frequency − and applying Z-score normalization along with principal component analysis, the system effectively extracts critical features to improve modeling efficiency. The proposed model employs a dual hidden-layer architecture, integrating the Swish activation function and the Rectified Adam optimizer to balance convergence speed and training stability. A hybrid data strategy is adopted during training, combining public datasets with empirical measurements. Key hyperparameters are optimized through grid search. Experimental results show that the model outperforms mainstream methods such as Long Short-Term Memory networks and eXtreme Gradient Boosting in fatigue state classification. It achieves superior accuracy (89.4%) and area under the receiver operating characteristic curve (0.93), along with a 22.2% reduction in mean squared error. In high-fatigue scenarios, the model attains an accuracy of 90.1%, demonstrating strong sensitivity and specificity. This study presents a novel approach to physiological signal modeling and contributes to the advancement of deep learning applications in sports science. It offers both theoretical insights and practical support for the development of intelligent training systems. [ABSTRACT FROM AUTHOR]