Chinese Small Business Credit Scoring: Application of Multiple Hybrids Neural Network

Abstract

In recent years, hybrid models have proven to be a promising approach for the forecasting of credit status, therefore, the aim of this project is to examine the prediction performance of hybrid classifiers. Particularly, the combination of the feature engineering with popular neural network (NN) classifiers; an hybridization approach, is compared with hybrid classifier, NN classifiers, and three well-known baseline classifiers, i.e. stepwise discriminant analysis (SDA), stepwise logistic regression (SLR), and decision trees (DTs). Overall, we executed a 12+8+ (8×8) experimental design that resulted in 84 unique classification models; i.e., 12 baseline models, 8 NN models, and 64 hybrid models, a multiple hybrid; are examined over a large credit scoring dataset from a Chinese commercial bank. Besides, thirteen evaluation measures are used for the assessment task and this may be the first effort to link up multiple hybrid classifiers with multiple performance metrics for the evaluation of small business credit. The results reveal that the predictive and distinguish ability of the F ratio based SDA with multilayer perceptron based NN classifier (SDA FR +MLP), a hybrid model, outperforms both of the one-dimensional scoring models (baseline model and NN model) and its hybrid counterparts.