샘플링 불확실성 하에서 신축망에 대해 부트스트랩 방법을 이용한 입력변수의 유의성 분석

초록

The problem of high-dimensional input variables can be solved by analyzing the significance of input variables using an elastic net. However, the significance varies because of sampling uncertainty, which can lead to incorrect inferences. Specifically, the sampling uncertainty tends to increase as the dimension of input variables increases. Therefore, a significance analysis method using bootstrap method for the elastic net is presented herein to reduce the sampling uncertainty. Through a mathematical example, the proposed significance analysis method was confirmed to provide greater accuracy than the elastic net. Additionally, the significance of input variables was analyzed by applying the proposed significance analysis method to an engineering problem. Although the bootstrap method entails high computational costs, it is expected that meaningful results can be obtained at a reasonable cost.

키워드