Construction and empirical study of a multi-source data fusion model for adolescent health literacy assessment

To overcome limitations of traditional adolescent health literacy assessment -single data source, insufficient accuracy and interpretability -a multi-source data fusion model is developed.This study has four key phases: first, extract and combine features from four data types (questionnaire, physiological sensor, behaviour record, environment); second, propose a deep learning model integrating a two-path feature encoder and random forest to enhance fitting ability; third, use SHAP and LIME for global and local interpretability analysis to assess key feature weight changes across groups; finally, verify model performance through multiple empirical experiments.Experiments show the integrated model achieves an F1-score of 0.873 and test-set AUC of 0.921, outperforming comparative models significantly while maintaining stable performance across age and gender heterogeneous groups.This paper verifies the validity of multiple data fusion strategies and provides an extensible algorithm framework with an interpretable mechanism.