CAJ | 학술논문

采用模拟研究的方法，分别在回归预测和分类判别两种环境中讨论有监督Group MCP方法在不同结构错误率下进行变量选择和结果预测的稳健性，并通过实例分析讨论本研究的实用价值。研究结果显示：忽略解释变量的内部结构进行变量选择会导致很多重要解释变量被疏漏，而有监督Group MCP方法考虑了解释变量的内部结构，在结构错误率低于5％时会以不低于98％的概率选出有效解释变量，并尽量降低冗余变量被选择的可能性。此研究成果为有监督Group MCP方法的合理使用奠定了基础。
채용모의연구적방법，분별재회귀예측화분류판별량충배경중토론유감독Group MCP방법재불동결구착오솔하진행변량선택화결과예측적은건성，병통과실례분석토론본연구적실용개치。연구결과현시：홀략해석변량적내부결구진행변량선택회도치흔다중요해석변량피소루，이유감독Group MCP방법고필료해석변량적내부결구，재결구착오솔저우5％시회이불저우98％적개솔선출유효해석변량，병진량강저용여변량피선택적가능성。차연구성과위유감독Group MCP방법적합리사용전정료기출。
T he structure got from either the qualitative or the quantitative method is an approximation of the real one ,and has a structural error rate .So whether or not the supervised Group MCP having robustness property is an urgent problem .In this paper ,we simulate sets of data with different structure error rates in both regressions and classifications and analyze them by supervised Group MCP to discuss Group MCP method's robustness in selecting variables and model's predictions . And then a practical application is used to test this study's pragmatic value .These results show that selecting variables without considering the latent structure of the explaining variables will result in the omission of significant features ,and that Supervised Group MCP takes the explaining variables'structures into consideration and evidently selects the important variables and reduce the probability of selecting redundant variables w hen the structural error rate is lower than 5% ,which lay foundations for supervised Group MCP method's appropriate use .