Identifying the optimal predictors for adolescent mental and physical health using machine learning methods
Sun, Y., Lin, Y., Ho, H. C. Y., Lu, J., Tse, C. Y., Gao, X., Liu, H., Poon, K.-T., Jiang, D., Li. L. M. W.* (in press). Identifying the optimal predictors for adolescent mental and physical health using machine learning methods. Journal of Affective Disorders. https://doi.org/10.1016/j.jad.2025.121046
2024 Impact Factor 4.9
2024 JCR Rank 33/288, Q1 in Psychiatry
Abstract
Background
Adolescent health is a worldwide concern. Previous research identified many factors of mental and physical health issues, which typically focused on a single or a few domains of predictors.
Methods
To identify the optimal predictors for adolescent mental and physical health, this study adopted multiple machine learning methods to examine the effect of nine-domain 46 predictors in predicting mental and physical health among adolescents from Hong Kong, China, and the Netherlands using data from PISA, 2022.
Results
The results showed that, of all predictors considered, the emotional and social capital domains were most crucial for adolescent mental and physical health in both societies. The results also revealed nuanced patterns in the two societies, with the environmental domain being more critical in Hong Kong and the physical domain being more critical in the Netherlands. The model trained based on the data of one society performed worse in predicting health outcomes in another society.
Conclusion
The results suggest that interventions targeting emotion and social capital domains may be useful in addressing mental and physical health issues among adolescents. However, culturally specific strategies should be considered for interventions, as the results highlight the importance of considering cultural backgrounds despite the overlapping patterns observed in the two societies.