Multi-domain physical fitness profiles and machine learning classification by sport category across 13 competitive sports

Background: Comparative multi-sport fitness profiling using a common institutional assessment framework remains limited in Southeast Asian competitive sport.

Aims: To compare multi-domain fitness profiles across 13 sports grouped into four sport categories and to evaluate machine-learning classification of sport category.

Methods: This cross-sectional STROBE-reported study included 197 KONI athletes (137 males, 60 females; age 20.6 ± 5.8 years). Standardized procedures were used within applicable sport groups, although several tests were category-dependent. Athletes were grouped as Combat (n=69), Team (n=72), Individual Technical (n=37), and Precision/Shooting (n=19). Between-category analyses and Random Forest (RF) and Linear Discriminant Analysis (LDA) with 5-fold out-of-fold cross-validation were performed.

Results: The largest between-category effects were observed for vertical jump (η²=0.438, p<.001), the Bleep Composite/estimated VO₂max signal (η²=0.217, p<.001), and sit-up endurance (η²=0.199, p<.001). RF achieved 79.7% accuracy and macro-AUC=0.907; the three largest Gini importances in the final nine-variable model were vertical jump (0.211), Bleep Composite (0.162), and push-up endurance (0.144).

Conclusion: Multi-domain fitness patterns differed across the four administrative sport categories, but age, maturation, sex composition, and category-linked test availability constrain interpretation. The classifier is exploratory and requires external validation; the

Findings: describe already-trained athletes and do not establish prospective talent-identification validity.