This study introduces a machine learning method (DML) recently used for causal analyses on labor market issues and presents the results of applying DML to Korea’s vocational training data. Voluminous studies have estimated the effect of training on employment and they have come up short of consensus. The DML’s flexibility in setting relationships of variables may help to avoid problems related to model specification. The DML’s properties of the debiasedness and robustness may also provide what the reasonable range of training effects should be. In addition, the paper attempt heterogeneous treatment effects, which will be informative to policy implementation.