Satisfactory Feature Selection andItsApplication in Enterprise Credit Assessment

Theselection ofevaluating index system isoneofthe keyproblems inenterprise credit assessment. Itisessentially a satisfactory feature selection (SFS)problem. In thispaper, several novelsatisfactory-rate functions offeature set(SRFFS) aredesigned, inwhichtheclassification performance ofthe feature subset anditssize areconsidered compromisingly. The accuracy ofSVM CrossValidation isemployed asevaluation criterion ofclassification ability, andtheSFSalgorithm is described indetail. Contrastive experiments arecarried onSFS andthreeotherdifferent feature selection methods:S-SFS, Expert+GAFS andGAFS.Results showthatSFS,whichcanpick outthefeature subset withlowdimension, highclassification accuracy andbalanced ranking performance, issuperior tothree other ones. Keywords-enterprise credit assessmen; feature selection; satisfactory optimization; support vector machine (SVM)