Integrating machine learning and statistical analysis to forecast insufficient physical activity trends using socio-demographic predictors

Introduction Insufficient physical activity is a major global health issue with disparities across sex and regions. This study examines these inequalities and projects future trends. Methods World Health Organization data were analyzed using a Linear Mixed-Effects Model to assess the effects of sex, time, and region, accounting for country-level variation. Geospatial analysis was conducted, and four machine learning models (XGBoost, Random Forest, Multi-Layer Perceptron, Support Vector Regression) were used for forecasting. Results Females showed higher and more variable levels of insufficient physical activity than males. Prevalence was lower in Africa and South-East Asia and higher in the Americas and Eastern Mediterranean. Trends worsened over time, particularly among females and high-prevalence regions. Forecasts differed: tree-based models suggested stabilization, while neural and kernel-based models projected continued increases. Discussion Persistent disparities by sex and region remain. Variation across models highlights uncertainty in projections, supporting ensemble approaches and the need for targeted, gender-responsive interventions.