Prediction of adolescent weight status by machine learning: a population-based study

Area of Science:

• Public Health
• Machine Learning
• Adolescent Health

Background:

• Adolescent weight problems are a growing public health concern.
• Early prediction of non-normal weight status is crucial for prevention.
• Few temporal prediction tools for adolescent weight status exist.

Purpose of the Study:

• To predict the short- and long-term weight status of Hong Kong adolescents.
• To assess the importance of various predictors for adolescent weight status.

Main Methods:

• A population-based retrospective cohort study using health assessment data from Hong Kong adolescents.
• Six prediction models (Decision Tree, Random Forest, k-NN, XGBoost, SVM, logistic regression) were generated using diet, physical activity, psychological well-being, and demographics.
• Model performance was evaluated using standard classifier metrics and Shapley values for predictor importance.

Main Results:

• The eXtreme gradient boosting (XGBoost) model demonstrated superior performance in predicting long-term weight status.
• The XGBoost model achieved high accuracy (0.72-0.74) and AUC values (0.83-0.93) for predicting weight status.
• Weight, height, sex, age, and aerobic exercise frequency/duration were key predictors.

Conclusions:

• Machine learning models accurately predict adolescent weight status in both short and long term.
• The developed multiclass model enables accurate long-term prediction using easily assessed variables for self-prediction.
• Interpretable models can guide early, individualized interventions for adolescents with weight concerns.