Abstract

Many studies for youth unemployment rely on data from the same survey year or outdated sources, limiting their ability to capture recent shifts in the labor market. Furthermore, existing research often employs conservative models with a narrow set of variables or fails to effectively address incomplete data caused by the hierarchical structure of survey questionnaires, resulting in reduced predictive accuracy and interpretability. To address these limitations, this study utilizes data from the 2021 Youth Panel Survey to predict the employment status of unemployed youth one year later, reflecting the latest labor market trends. We apply the Generalized, Unbiased, Interaction Detection and Estimation(GUIDE), which is well-suited for handling incomplete data, to ensure both interpretability and predictive accuracy. Comparisons with ensemble models, such as Random Forest, XGBoost, and CatBoost, demonstrate that GUIDE performs competitively, offering superior interpretability. Additionally, the tree estimated by GUIDE provides meaningful insights into how key variables–including the job search activities and employment preparation before and after graduation, preparation for government exams, existence of first job after graduation, job search activities during employment preparation–impact employment outcomes after one year.


http://dx.doi.org/10.5351/KJAS.2025.38.3.357