자동화 머신러닝(AutoML)과 SHAP을 적용한 청소년의 다중시민성 유형 및 설명 변인 탐색

Abstract

This study aims to investigate the typology of multiple citizenship among Korean adolescents and the predictive factors in the global era. The study analyzed data from the 6th wave of the 2013 Korean Educational Longitudinal Study and utilized latent profile analysis to explore the typology of multiple citizenship. To identify the predictive factors determining multiple citizenship, an automated machine learning algorithm was used to find the most performant model. SHAP (Shapley additive explanations) techniques were then applied to provide interpretable explanations of the machine learning results. The key findings are as follows. First, in the latent profile analysis, the multiple citizenship of adolescents was categorized into “high multiple citizenship,” “moderate multiple citizenship,” and “low multiple citizenship.” Second, the results of automated machine learning showed that gradient boosting exhibited the highest performance among machine learning techniques. When predicting the key variables related to adolescents’ multiple citizenship through machine learning, it was found that communication competence emerged as the most important predictor, followed by cognitive information processing skills, awareness of cultural and artistic activities, independent career maturity, educational experiences involving cultural diversity and international issues, gender, and relationships with teachers and peers. Based on these results, this study provides policy recommendations and implications for effective civic education for adolescents.