Despite sustained government efforts to support multicultural students in Korea, little empirical evidence exists on whether a multicultural background actually causes lower academic achievement. This paper applies the Double/Debiased Machine Learning (DML) framework of Chernozhukov et al. (2018) to three panel cohorts from the Korean Children and Youth Panel Survey 2010. Multiple machine learning methods—including Lasso, Ridge, Random Forest, Boosted Trees, and Neural Networks—flexibly control for socioeconomic and educational covariates, while Neyman-orthogonal score functions and cross-fitting ensure robustness to regularization bias and functional form misspecification. Our principal finding is that, after controlling for observed socioeconomic and educational covariates, there is no statistically robust evidence that multicultural background lowers academic achievement independently of these factors. For the Grade 1 and Grade 4 cohorts, the raw achievement gap is no longer statistically distinguishable from zero once covariates are accounted for. For the Grade 7 cohort, some specifications yield marginally significant negative effects, but these are not robust across learners or ensemble methods. These results suggest that the observed gap is driven by socioeconomic factors rather than multicultural status itself, calling for a reassessment of policies that target academic support exclusively at multicultural students.