Journal of Risk Model Validation

Risk.net

An optimized support vector machine intelligent technique using optimized feature selection methods: evidence from Chinese credit approval data

Mohammad Zoynul Abedin, Chi Guotai, Fahmida - E - Moula, Tong Zhang and M. Kabir Hassan

  • Data splitting plays a major role in assessing the true performance of the models.
  • Average results from “sample division example set” achieved the more robust prediction ability than “no sample division” dataset
  • Selecting significant features extensively improve the credit approval accuracy.
  • ‘The business cycle index,’ X11 is the most contributed features. 

This paper focuses on feature selection methods for support vector machine (SVM) classifiers, checking their optimality by comparing them with some statistical and baseline methods. To achieve the above objective, we exploit twelve feature selection methods from the family of filters and embedded approaches by splitting a Chinese database. Our findings suggest that the average result from sample division cases will achieve a more robust prediction ability than that from “no sample division” cases. Moreover, ridge regression (SVM9) in training and “average results from sample division” data sets, along with DTQUEST (SVM7) in “no sample division” example sets, give outstanding performance with respect to all performance criteria. With these contributions, therefore, our paper complements previous evidence and modernizes the methods of feature selection to render SVM classifiers favorable for credit approval data modeling. This study has practical implications for financial institutions, managers, employees, investors and government officials looking to sort out forthcoming lending transactions to attain a target risk/return trade-off.

To continue reading...

You need to sign in to use this feature. If you don’t have a Risk.net account, please register for a trial.

Sign in
You are currently on corporate access.

To use this feature you will need an individual account. If you have one already please sign in.

Sign in.

Alternatively you can request an indvidual account here: