KMAN Publication Inc. (KMANPUB) Journal of Adolescent and Youth Psychological Studies (JAYPS) 2981-2526 7 Serial Number 45 2026 05 01 Explainable Deep Learning Models of Adolescent Eating Disorders Based on Appearance Comparison, Influencer Exposure, Emotion Dysregulation, Self-Criticism, and Weight Stigma Explainable Deep Learning Models of Adolescent Eating Disorders Based on Appearance Comparison, Influencer Exposure, Emotion Dysregulation, Self-Criticism, and Weight Stigma 1 14 10.61838/kman.jayps.5457 EN Journal Article 2025 09 11 <table> <tbody> <tr> <td> <p><strong>Objective:</strong> The present study aimed to develop and interpret explainable deep learning models for predicting adolescent eating disorder symptoms based on appearance comparison, influencer exposure, emotion dysregulation, self-criticism, and weight stigma.</p> <p><strong>Methods and Materials:</strong> This study employed a cross-sectional predictive design using explainable artificial intelligence approaches among 1,219 Canadian adolescents aged 14 to 18 years recruited from secondary schools in Ontario, British Columbia, and Quebec. Participants completed standardized measures assessing eating disorder symptoms, appearance comparison, influencer exposure, emotion dysregulation, self-criticism, and weight stigma through an online survey platform. Data were analyzed using TensorFlow, Keras, and Scikit-learn libraries. The dataset was divided into training, validation, and testing subsets, and a multilayer explainable deep neural network was developed. SHapley Additive exPlanations (SHAP) and Local Interpretable Model-Agnostic Explanations (LIME) were used to enhance model interpretability and identify the relative contribution of predictors. Model performance was compared with logistic regression, support vector machine, and random forest algorithms.</p> <p><strong>Findings: </strong>Significant positive associations were observed among all study variables and eating disorder symptoms. Self-criticism demonstrated the strongest association with eating pathology, followed by emotion dysregulation and weight stigma. The explainable deep learning model achieved the highest predictive performance compared to alternative machine learning models, with superior accuracy, precision, recall, F1-score, and area under the receiver operating characteristic curve values. SHAP analyses identified self-criticism as the strongest predictor, followed by emotion dysregulation, weight stigma, appearance comparison, and influencer exposure. Interaction analyses further demonstrated that the predictive effects of appearance comparison and weight stigma intensified when combined with elevated self-criticism and emotional dysregulation.</p> <p><strong>Conclusion:</strong> The findings suggest that adolescent eating disorders are shaped by complex interactions among emotional vulnerabilities, maladaptive self-evaluative processes, and digitally reinforced appearance pressures. Explainable deep learning models provide a powerful and interpretable framework for identifying multidimensional psychological risk patterns associated with eating pathology and may support more personalized prevention and intervention strategies for adolescents.</p> </td> </tr> </tbody> </table> https://www.journals.kmanpub.com/index.php/jayps/article/download/5457/9877