Prediction of Social Isolation and Vulnerability Factors among Isolated and Withdrawn Youth Using Machine Learning

Objectives: This study aimed to predict social isolation among youth using machine learning and to identify key vulnerability factors, thereby providing empirical evidence for early detection and preventive policy. Methods: Data were obtained from the 2022 Survey on Socially Isolated and Withdrawn Youth conducted by the Seoul Metropolitan Government. A total of 5,353 individuals aged 19–39 years were included in the analysis. Social isolation was defined as meeting both emotional and physical isolation criteria. Four machine learning classification models—k-nearest neighbors, Naive Bayes, decision tree, and random forest—were applied. Model performance was evaluated using accuracy, precision, recall, F1-score, and the area under the receiver operating characteristic curve (ROC-AUC). Feature importance was examined primarily using tree-based models. Results: Among the four models, the random forest model demonstrated the most stable and balanced predictive performance, showing high accuracy and AUC values. Key predictors of social isolation included frequency of going out, number of recent social interactions, household socioeconomic status, subjective health status, and employment-related factors. Social relationship and interaction variables consistently showed the strongest influence on isolation risk. Conclusions: Machine learning–based analysis is an effective approach for identifying youth at risk of social isolation. The findings suggest that behavioral and socioeconomic indicators can support early screening systems and contribute to data-driven policy strategies for preventing prolonged social isolation among youth.