Journal of Risk Model Validation

Risk.net

An advanced hybrid classification technique for credit risk evaluation

Chong Wu, Dekun Gao, Qianqun Ma, Qi Wang and Yu Lu

  • We find a new hybrid classifier based in DBN to credit evaluation.
  • Our research finds the new model is better than some traditional base classifiers with a German and Australian credit dataset.
  • The model has a better performance in the imbalanced German dataset than balanced Australian dataset.

Due to the important role that credit risk evaluation (CRE) plays in banks, it is crucial that we improve the accuracy of the CRE model. In this paper, we employ a hybrid approach to design a practical and effective CRE model based on a deep belief network (DBN) and the K-means method. First, using the K-means method, we preprocess the original data set and reclassify the unrepresentative data into high grades. Second, based on the results of our K-means method analysis, a DBN is used to construct a CRE model. We randomly select data in fixed proportions from the samples that have been assigned to new classes in order to train and test the DBN model. Finally, German and Australian credit data sets are applied to the proposed model. Our results show that the hybrid classifier we propose is effective in CRE and performs significantly better than classical CRE models.

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 individual account here: