Satisfactory Feature Selection and Its Application in Enterprise Credit Assessment

The selection of evaluating index system is one of the key problems in enterprise credit assessment. It is essentially a satisfactory feature selection (SFS) problem. In this paper, several novel satisfactory-rate functions of feature set (SRFFS) are designed, in which the classification performance of the feature subset and its size are considered compromisingly. The accuracy of SVM cross validation is employed as evaluation criterion of classification ability, and the SFS algorithm is described in detail. Contrastive experiments are carried on SFS and three other different feature selection methods: S-SFS, Expert+GAFS and GAFS. Results show that SFS, which can pick out the feature subset with low dimension, high classification accuracy and balanced ranking performance, is superior to three other ones.