Predicting Adolescent Concern Over Unhealthy Food Ads Using Explainable AI

Machine learning and explainable AI techniques are powerful mechanisms to analyze the influence of unhealthy food advertisements on health awareness and concern levels. This research aims to predict levels of concern about unhealthy food advertisements and identify significant demographic and behavioral variables that modify the need for these concerns. A survey was conducted, and responses from 1030 adolescents were gathered on their knowledge about food advertisements and what affects their perceptions of health. Stacking Ensemble, Random Forest, and XGBoost with hyperparameter optimization and synthetic oversampling techniques were applied to predict three categories of concern levels in unhealthy food advertisements. Stacking Ensemble, utilizing Random Forest, K-Nearest Neighbors, and Support Vector Machines as base learners and Logistic Regression as the meta-learner, yielded the highest prediction accuracy of 98.54% and an F1 score of 0.99. The most relevant features for predicting the food advertisement concern levels were age, parental education and advertisement types, such as free toys and celebrity endorsement. LIME and SHAP-integrated interpretability methods have been utilized for a proper understanding of how advertisement strategies influence adolescent health awareness. The developments resulted from growing research efforts in which predictive analytics is connected to explainability in understanding and mitigating the adverse influence of unhealthy food advertisements. The results bear implications for public health policies, advertising regulations, and educational campaigns moving against improved adolescent dietary behaviors and awareness. The private dataset and implementation codes are at the following link: https://github.com/MdFahimShahoriar/adolescent-food-ad-concern-prediction