A lifestyle-based prediction model for obesity in Chinese adolescent students

Adolescent obesity has emerged as a critical global public health challenge, necessitating effective tools for early identification and intervention. This study aimed to identify significant contributing factors and develop a predictive model for adolescent obesity using machine learning algorithms.

An anonymised dataset of 2,338 adolescents was utilised, incorporating variables related to family factors, lifestyle behaviours, and physical fitness scores. Variable selection was performed using LASSO regression with k-fold cross-validation, followed by parameter estimation via logistic regression. The optimal classification threshold was determined using the Youden Index.

The final predictors included gender, mother's educational level, parental BMI, weight at age 12, parenting style, weekly sweets consumption frequency, meal duration, sleep duration, and physical fitness score. The model demonstrated robust performance, with an AUC of 0.91, accuracy of 0.86, and sensitivity of 0.84. Subgroup analysis indicated consistent performance across genders, with slightly superior predictive efficacy in males (AUC = 0.912) compared to females (AUC = 0.898).

The proposed interpretable framework combines high predictive accuracy and sensitivity, offering a valuable tool for timely identification and intervention in high-risk adolescents. These findings underscore the potential of data-driven approaches in addressing adolescent obesity.