ScholarWorks@중앙대학교

초록

This study aims to examine the current state of youth runaway in South Korea, identify limitations in existing research, and propose a new analytical approach to address these gaps. Prior studies on youth runaway behavior have relied heavily on regression-based methods, which restrict the number of variables that can be considered due to issues such as multicollinearity and overfitting. As a result, findings across studies have often lacked consistency, and the identification of policy-relevant predictive factors has been limited. To overcome these challenges, this study applies Random Forest, a machine learning technique, to simultaneously incorporate a large set of variables and to identify the most influential predictors of youth runaway behavior. The analysis reveals that the key predictors differ between runaway impulses and actual runaway behavior. While runaway impulses are most strongly predicted by feelings of isolation—such as loneliness, lack of trusted confidants, and communication breakdown with parents—actual runaway behavior is more strongly associated with delinquent activities such as smoking and drinking, although emotional factors like loneliness also play a meaningful role. Notably, this study highlights the significant predictive power of loneliness in both runaway impulses and runaway behavior, a factor that has received limited attention in previous research on determinants of youth runaway behavior, thereby offering an important academic contribution.

키워드