Unsupervised learning-based optimization of college students' fundamental capacity evaluation method.

Background: The Fundamental Capacity Screen (FCS) is a standardized method for assessing athletic performance and fundamental capacity (FC). However, its current framework lacks refinement, yielding qualitative and insufficiently detailed results that hinder precise athletic training and performance evaluation. Objective: This study employs unsupervised learning to refine FC evaluation, enhancing its objectivity and scientific rigor by providing a more data-driven and interpretable approach. Methods: Data from 123 college students (61 males, mean age 20.42 ± 1.88 years; 62 females, mean age 20.35 ± 1.85 years) regarding FC test indicators were initially subjected to a factor analysis (FA) model to identify underlying factors. Hierarchical clustering was then performed on the first factor. Based on these results, relevant rules of abilities were established, and principal component analysis (PCA) was employed to reduce the dimensionality of multidimensional indicators. Finally, the entropy weight method was used to calculate and adjust the scores of each ability, and the scores were summarized and classified by percentile. Results: FA revealed 3 distinct factors (jumping ability, upper limb ability, and lower limb ability) that together explained 0.8569 of the total variance. The hierarchical clustering dendrogram indicated that jumping ability could be further subdivided. Based on this, 5 ability categories were defined: upper limb balance control, lower limb balance control, core-mediated transfer control, explosive control, and impact control. The composite scores were categorized into 5 intervals: poor, pass, good, excellent, and outstanding. The depicted radar map shows the median characteristics of the composite scores of different abilities. Conclusions: Results demonstrate that unsupervised learning provides a more intuitive and objective framework for FC assessment, offering a novel perspective for athletic performance evaluation and training optimization.