关键词: 评价指标筛选；信息重叠；综合评价；反映象相关矩阵；取样适切性量数

Abstract: Evaluation is the basis of scientific decision-making and the most common cognitive activity of human beings. Evaluation index system is the basis of evaluation activities, so it is very important to select evaluation indicators scientifically. A group of evaluation indicators constitute an interactive and interactive system, and there are complicated information overlapping among them. Moreover, the overlapping information will be repeatedly emphasized in the comprehensive evaluation, which will distort the evaluation results. Therefore, it is necessary to eliminate some indicators with high information overlap to reduce the information overlap level of indicator set. The existing research mainly solves this problem by reducing the correlation between one indicator and some (mainly another) indicators. It is very easy to eliminate more indicators, but the information overlap of indicator set decreases slowly, which leads to excessive loss of evaluation information. In order to solve the above problems, Measure of Sampling Adequacy (MSA) is first introduced to represent the information overlapping level between an evaluation indicator and all other indicators, and KMO test statistics is introduced to represent the overall information overlapping level of a group of evaluation indicators. Secondly, the indicators with the largest MSA value in the indicator set are removed in turn until the KMO value of the remaining indicators is not high, which can quickly reduce the overall information overlap of the evaluation indicator set. After that, one with higher MSA value is eliminated in any two indicators with high partial correlation level to avoid the high information overlap between individual indicators. Finally, taking a small enterprise loan customer credit evaluation index screening of a commercial bank in China as an example, compared with the most widely used screening methods of information overlap indicators, it is used to illustrate the feasibility and validity of the proposed method. The results show that this method has higher efficiency of reducing information overlap of indicator set and less loss of evaluation information; it can neither delete the evaluation indicators with low information overlap level nor retain the evaluation indicators with high information overlap level. The study of the evaluation indicator selection methods is expanded, and a reference and guidance for solving the problem of reducing the high correlation among indicators is provided.

Key words: evaluation indicator screening; information overlapping; comprehensive assessment;anti-image correlation matrix; measure of sampling adequacy