To improve the scientific accuracy and precision of children's physical fitness evaluations, this study proposes a model that combines self-organizing maps (SOM) neural networks with cluster analysis. Existing evaluation methods often rely on traditional, single statistical analyses, which struggle to handle the complexity of high-dimensional, nonlinear data, resulting in a lack of precision and personalization. This study uses the SOM neural network to reduce the dimensionality of high-dimensional health data. Moreover, it integrates cluster analysis to categorize and analyze key physical fitness attributes, such as strength, flexibility, and endurance. Experimental results show that the proposed optimized model outperforms comparison models such as T-distributed stochastic neighbor embedding, density peak clustering, and deep embedded clustering in terms of performance. The accuracy for the strength dimension reaches 0.934, the F1 score is 0.862, and the area under the curve of receiver operating characteristic is 0.944. The silhouette coefficients for cluster analysis in strength, flexibility, and endurance dimensions are 0.655, 0.559, and 0.601, respectively, demonstrating good intra-class and inter-class distances. The proposed model enhances the comprehensive analysis of children's physical fitness and provides a scientific basis for personalized health interventions, making an important contribution to research in this field.
Keywords: Assessment of physical fitness of young children; Cluster analysis; Data dimensionality reduction; Evaluation of children’s physical fitness; Nonlinear data processing; Personalized health interventions; SOM neural network.
© 2025. The Author(s).