Study On PCA-MLP Credit Rating Model For Incomplete Data And Its Applications

In this thesis we establish a PCA-MLP credit rating model for incomplete data.The main works are as follows:1.By applying three methods on the German data,we conclude that Multi-Layer Perception Neural Network performs better than Logistic regression and support vector machine on bank credit data.2.There are always missing values in real bank credit data.We introduce an algorithm based on principal component analysis to fill the missing data.Experiments show that this algorithm can fill incomplete data with smaller error than several commonly used methods.3.Based on the above two conclusions,we establish a PCA-MLP credit rating model for incomplete data,in which we keep the ratios between good and bad costumers equal in training set and in test set.At last,we apply our model on credit data of a bank in Guangzhou and obtain a high-accuracy discriminant.