A data-driven stratification framework for physical fitness in junior high school students: a composite approach based on BMI and multidimensional data.

Objective: To overcome the practical challenges of implementing differentiated instruction in large physical education (PE) classes, this study develops a composite stratification framework. This exploratory framework provides a baseline for precise, individualized physical fitness interventions in adolescents. Methods: A quantitative cross-sectional design was employed. First, a questionnaire survey was administered to 125 primary and secondary school PE teachers in Beijing to evaluate their attitudes and preferences for evaluation indicators regarding differentiated instruction. Second, multidimensional physical fitness monitoring data were collected from 532 junior high school students. Following the standardization of continuous physical fitness test scores and categorization by Body Mass Index (BMI), one-way analysis of variance (ANOVA) and K-means clustering algorithms were comprehensively applied to explore the underlying latent cluster formations of students' multidimensional physical fitness characteristics. Results: The survey revealed a high theoretical acceptance of differentiated instruction among frontline teachers, yet a "knowing-doing gap" remains in large-class settings, with 65.85% of respondents prioritizing BMI as the primary stratification criterion. ANOVA confirmed significant differences in physical fitness scores across BMI groups, validating its preliminary screening value. However, it also demonstrated that relying on a single morphological indicator easily obscures the micro-heterogeneity of unbalanced physical performance. The introduction of multidimensional data clustering successfully grouped male students into 2 clusters and female students into 3 clusters, effectively controlling pedagogical stratification tiers and intra-group variance. Conclusion: The composite stratification framework-integrating initial BMI screening with multidimensional physical fitness data clustering-serves as an exploratory framework for identifying heterogeneous fitness profiles. Because this study identifies statistical subgroups cross-sectionally without prospective interventions or safety monitoring, it does not constitute a validated educational strategy. Its practical pedagogical effectiveness, longitudinal stability, and safety require rigorous external validation studies.