Metamodeling is a method that approximates the actual model and has been used in various fields to solve engineering problems with high computational cost. However, if the number of design variables increases, the number of samples required to satisfy a reasonable accuracy increases exponentially, making it difficult to apply. However, there is a lack of research on a practical metamodeling technique to solve this problem. In this study, a meta-modeling and sampling strategy that increases the efficiency of metamodel generation by decomposing design variables and creating sub-space to perform reliability analysis is proposed. This method is used to obtain optimal design of high-dimensional and high-cost simulation models. The numerical example shows that the proposed method shows that the efficiency is significantly improved compared to the conventional approach.