Distinguishing characteristics of out‐of‐school adolescents in South Korea: A machine learning approach

Abstract Recently in South Korea the increasing prevalence of school dropouts and the declining age at which students leave school have drawn renewed attention to this issue. In line with preventive efforts and recognizing early signs of leaving school, the current study aims to identify a set of variables that are most important for understanding the experience of school dropout among South Korean adolescents. Data from two independent panel studies collected by the National Youth Policy Institute in South Korea were merged and analyzed in this study: Korean Children and Youth Panel Study ( N = 1646, age = 15.90, girls = 50.73%) and Dropout Youth Panel Study ( N = 609, age = 16.84, girls = 56.16%). We applied machine learning algorithms to classify the experience of school dropout using two analytic methods: random forest and decision tree. A total of 36 features from personal, family, school, peer, and community domains were used in the analyses. Specifically, adolescent behavioral characteristics (truancy, smoking, drinking, media use), family structure, teacher relationship, group bullying victimization, and collective efficacy, were consistently identified as significant features of school dropout in random forest and decision tree models. Such information, which highlights a broad spectrum of important factors within adolescents' ecological systems, may provide a scientific knowledge base for school‐level prevention efforts. By identifying these features, social workers and educators may develop early warning systems against school dropouts and accurately screen adolescents with high risk.