KMAN Publication Inc. (KMANPUB) Journal of Adolescent and Youth Psychological Studies (JAYPS) 2981-2526 7 Serial Number 45 2026 05 01 A Multimodal Machine Learning Approach to Adolescent Loneliness Using Social Disconnection, Parasocial Relationships, Emotional Inhibition, Family Cohesion, and Digital Isolation A Multimodal Machine Learning Approach to Adolescent Loneliness Using Social Disconnection, Parasocial Relationships, Emotional Inhibition, Family Cohesion, and Digital Isolation 1 13 10.61838/kman.jayps.5464 EN Journal Article 2025 09 18 <table> <tbody> <tr> <td> <p><strong>Objective:</strong> The present study aimed to develop and evaluate a multimodal machine learning model for predicting adolescent loneliness based on social disconnection, parasocial relationships, emotional inhibition, family cohesion, and digital isolation among adolescents in the United States.</p> <p><strong>Methods and Materials: </strong>The study employed a cross-sectional predictive correlational design using multimodal machine learning techniques. The statistical population consisted of adolescents aged 13 to 18 years enrolled in secondary schools across the United States during the 2025–2026 academic year. Using stratified multistage sampling, 1,248 adolescents were recruited, of whom 1,182 participants were included in the final analysis after data screening. Data collection involved standardized psychological instruments assessing loneliness, social disconnection, parasocial relationships, emotional inhibition, and family cohesion, alongside objective digital behavioral indicators including screen time, passive scrolling duration, nocturnal device usage, and solitary online activity. Data were analyzed using Python-based machine learning libraries and included Random Forest, Support Vector Machine, XGBoost, Multilayer Perceptron Neural Network, and a multimodal ensemble learning model. Model evaluation indices included accuracy, precision, recall, F1-score, and AUC-ROC values. SHAP-based explainable artificial intelligence analysis was additionally employed to determine feature importance.</p> <p><strong>Findings: </strong>The findings indicated significant positive relationships between adolescent loneliness and social disconnection, emotional inhibition, parasocial relationships, and digital isolation, while family cohesion demonstrated a significant negative association with loneliness. The multimodal ensemble model achieved the highest predictive performance compared to all other machine learning models, with an accuracy of 0.94 and an AUC-ROC value of 0.97. SHAP feature importance analysis revealed that digital isolation was the strongest predictor of adolescent loneliness, followed by social disconnection, family cohesion, emotional inhibition, and parasocial relationships. Hierarchical regression analysis further demonstrated that the predictor variables collectively explained 71% of the variance in loneliness scores.</p> <p><strong>Conclusion: </strong>The findings suggest that adolescent loneliness is a multidimensional phenomenon shaped by the interaction of emotional, social, familial, and digital behavioral factors. Multimodal machine learning approaches integrating psychosocial measures with objective digital behavior indicators may provide highly accurate and interpretable predictive models for identifying adolescents at risk for loneliness. Strengthening family cohesion, improving emotional communication, and promoting healthier patterns of digital engagement may contribute significantly to reducing loneliness and improving adolescent mental health outcomes.</p> </td> </tr> </tbody> </table> https://journals.kmanpub.com/index.php/jayps/article/download/5464/9879