부트스트랩 신뢰구간을 이용한 설계변수의 유의도에대한 통계적 분석

Abstract

As the number of design variables increases, convergence rate of optimization, the number of function calls, and design change cost can be worsened. To reduce the number of design variables, feature selection methods are being used, which analyze design variables that have a significant influence on responses. Since the feature selection methods use a trained model based on data, the significance of design variables can be varied due to the number, quality, and combination of the data. Therefore, this study analyzes the significance of design variables by adopting bootstrap method. Bootstrap resampling can generate a large number of bootstrap data, and based on this, bootstrap confidence intervals can represent a range of significance of design variables with probability. Lastly, instead of selecting the significant design variables based on the deterministic value of significance, the decision making is performed by considering the lower limit, upper limit, and length of the confidence interval.