Prediction of quality of life parameters among university students for the optimization of health-enhancing recreational programs

Abstract

Purpose. The present study sought to construct and empirically test predictive models of healthrelated quality of life (HRQoL) in a cohort of Ukrainian university students. The models incorporated behavioural patterns, psychosocial factors, and safety-related perceptions as key determinants. By doing so, the research aimed to generate evidence-based insights that could inform the development of more effective, context-sensitive recreational and health-promotion programs tailored to the challenges of studying and living under wartime conditions. Material & Methods. The study involved 172 students (48.8% male, 51.2% female; aged 18–25 years) enrolled in Ukrainian higher education institutions. Data collection included standardized questionnaires: the SF-36 Health Survey (Physical Component Summary (PCS) and Mental Component Summary (MCS)), the International Physical Activity Questionnaire – Short Form (IPAQ-SF), the Satisfaction With Life Scale (SWLS), and the Service Quality Assessment Scale (SQAS) adapted to wartime conditions. Factor analysis identified two main domains («Motivation» and «Constraints»), and additional variables such as time and financial limitations were analysed separately. Predictive modelling was conducted using Gradient Boosting Trees (GBT) and Random Forest (RF) methods. Model performance was evaluated with MAE, RMSE, and MAPE metrics. Results. The models demonstrated distinct predictive capacities for physical and mental components of HRQoL. PCS was effectively predicted using a simple GBT structure with high accuracy (MAE=5.53, RMSE=45.70, MAPE=0.102), while MCS required more complex modelling (130 trees). RF outperformed GBT for MCS prediction (RMSE=83.96 vs. 98.02; MAPE=0.180 vs. 0.198). Variable importance analysis revealed that safety was the strongest predictor of PCS, while life satisfaction (SWLS) and physical activity (IPAQ) were the most influential for MCS. In both models, sex and training frequency were the least significant predictors. Conclusions. Machine learning approaches provide valuable tools for predicting HRQoL among students in challenging contexts. The findings confirm that physical well-being is strongly influenced by environmental safety, whereas mental well-being depends more on subjective life satisfaction and physical activity. These results highlight the need for integrated healthenhancing recreational programs combining physical activity with psychosocial support and safety measures in Ukrainian universities during wartime.