Random forest-based identification and ranking of predictive factors for physical activity in Chinese college students

Bibliographic Data

ID22082245
AuthorsDing-you Zhang (0000-0003-4387-8799, Nantong University), Hu Lou (0000-0002-5617-6940, Nantong University), Jun Liu (0000-0001-5393-3187, Nantong University, corresponding author), Bo Li (0000-0002-7801-0319, Nantong University, corresponding author)
Year2025
Volume13
Pages1699606-1699606
Publication date2025-11-20
Open AccessYes
TypeARTICLE
VenueFrontiers in Public Health (JOURNAL)
Journal identifiersISSN: 2296-2565 • E-ISSN: 2296-2565
PublisherFrontiers Media SA (PUBLISHER • CH)
DOI10.3389/fpubh.2025.1699606
PMID41358216
OpenAlexW4416387381
LanguageEN
References cited78

Objective To explore the key predictors of physical activity (PA) levels of Chinese university students, and to analyse the predictive roles of different variables and their relative importance by means of the Random Forest (RF) algorithm. Methods A cross-sectional study was conducted using a stratified whole-group sampling method, covering 17 provinces of the country and collecting 10,182 valid questionnaires. Assessment of PA levels using the Physical Activity Rating Scale-3 (PARS-3) divides participants into attainment and non-attainment groups. The independent variables encompass the individual and interpersonal organisational levels of the socio-ecological model (SEM), comprising a total of 39 variables. These variables include demographic characteristics, psycho-behavioural factors, and social support, which were measured using several standardised scales. Feature importance analysis was performed using the Random Forest algorithm, and the model parameters were optimised with a grid search and 5-fold cross-validation to identify the most significant factors predicting PA. Results The RF model had an accuracy of 0.704 and an AUC value of 0.762. Characteristic importance analysis revealed that exercise adherence (exercise behaviour), sex, exercise adherence (effort investment), mastery of sports skills, exercise motivation (ability), alcohol consumption level, exercise adherence [emotional experience, exercise motivation (social), and exercise motivation (fun) ranked as the top nine predictive factors]. Specifically, all sub-dimensions of exercise adherence (exercise behaviour) positively predict PA (SHAP values > 0); sex, males are more likely than females to meet the standard group criteria (OR > 1, p < 0.001); mastery of sports skills correlates positively with PA levels; and among alcohol consumption level, ‘occasional drinking’ shows a negative correlation with the standard attainment rate ( p < 0.001). Conclusion Exercise adherence, sex, mastery of sports skills, and alcohol consumption level are significant factors predicting PA levels among Chinese university students. Recommendations for promoting PA include enhancing the “emotional value” and social attributes of exercise, addressing female students’ willingness to participate, and improving physical capabilities through skills training to effectively elevate activity levels

Alcohol consumption · Physical activity · Random forest · Motivation and Self-Concept in Sports · Physical Activity and Health · Physical Education and Training Studies

Tools

Open DOIOpen Access