Satisfactory Feature Selection Based On SVM And Its Application In Enterprise Credit Assessment

Credit assessment is the key step in loan business of commercial banks. The rationality and reliability of credit assessment will greatly affect the achievements of a bank.The commercial banks in china need some much better assessment methods to improve their competition ability, in which, nowadays, the traditional ratio-analysis method is popularly adopted. Recently, the intelligent models, such as Neural Networks (NN) and Support Vector Machines (SVM), have been introduced into the credit assessment domain and have already achieved some promising results. The existing researches are mainly focused on the establishment of the intelligent assessment models, but the optimal selection of the evaluating index system has still been rarely studied, In this paper, a new feature selection scheme, based on SVM, is adopted to perform the task of selecting the evaluating indexes. Furthermore, new modeling method is developed based on this scheme.We have designed the Satisfactory Feature Selection (SFS) scheme by applying satisfactory optimization technique into feature selection method. In SFS, the classification performance of the selected feature-subset and its size are considered compromisingly based on the characteristic of the credit data in domestic banks. Three other different feature selection schemes are presented to contrast with SFS. The experiment results show that SFS is superior to three other ones in the qualities such as the size of the selected subset, the predict performance on the total sample, and the predict performances equilibrium within the individual classes. A terse and reasonable evaluating index system is presented after the contrastive experiments.The ensemble methods of multi-classification SVMs are also studied in this paper. We manipulate the input feature-set to construct the classifier-ensemble, and create base-classifiers with large diversity by SFS. The contrastive experiments show that the ensemble based on SFS is superior to the ensemble based on bagging method, the ensemble based on feature grouping method and the ensembles based on three other feature selection methods. The multi-classification SVMs ensemble based on SFS also performances much better than a single SVM classifier.