KMAN Publication Inc. (KMANPUB) Journal of Adolescent and Youth Psychological Studies (JAYPS) 2981-2526 7 Serial Number 41 2026 01 10 Interpretable Deep Learning Analysis of Online Social Anxiety in Adolescents Using Feature Saliency Mapping Interpretable Deep Learning Analysis of Online Social Anxiety in Adolescents Using Feature Saliency Mapping 1 10 10.61838/kman.jayps.4996 EN Journal Article 2025 09 25 <table> <tbody> <tr> <td> <p><strong>Objective:</strong> The objective of this study was to develop and validate an interpretable deep learning model for predicting adolescents’ online social anxiety and to identify the most influential psychological and digital behavioral features contributing to its development.</p> <p><strong>Methods and Materials: </strong>A cross-sectional predictive modeling design was employed with a sample of 523 adolescents aged 13–18 years recruited from secondary schools in Austria. Participants completed standardized measures assessing online social anxiety, general social anxiety, self-esteem, perceived social support, emotion regulation, and digital behavior patterns. Data were analyzed using a deep neural network with multiple hidden layers and regularization techniques. Model performance was evaluated using RMSE, MAE, R², and Pearson correlation. Interpretability was achieved through advanced feature saliency mapping, including SHAP values and gradient-based attribution methods, enabling identification of both global and individual-level predictors of online social anxiety.</p> <p><strong>Findings: </strong>The deep learning model demonstrated strong predictive performance, explaining 81% of the variance in online social anxiety (R² = .81) with high correspondence between predicted and observed scores (r = .90). Feature saliency analysis revealed that online social comparison exerted the strongest positive influence on anxiety predictions, followed by low self-esteem, emotion dysregulation, general social anxiety, and daily social media use. Perceived social support displayed a robust negative contribution, functioning as a protective factor. Subgroup analysis indicated that adolescents with the highest anxiety levels exhibited intensified contributions from social comparison, emotional instability, excessive media engagement, and negative feedback sensitivity.</p> <p><strong>Conclusion: </strong>Interpretable deep learning provides a powerful and transparent framework for understanding the complex psychological mechanisms underlying adolescents’ online social anxiety, offering critical insights for early identification and targeted intervention.</p> </td> </tr> </tbody> </table> https://journals.kmanpub.com/index.php/jayps/article/download/4996/9005