Personal Credit Rating Based On The Neural Network Model Of Mutual Information Dimensionality Reduction

In China’s new normal economy, consumption increasingly drives the increasing of GDP. what accompanies with consumption stimulation is the expansion of personal credit and loan business, especially the unprecedented network credit and loan development represented by P2 P. The risks that come along with it have drawn more and more attention, and the most important one is credit risk. To scientifically quantify personal credit and establish comprehensive personal credit rating system is imperative for P2 P and commercial banks to control credit risk and increase returns.Currently, the research of domestic personal credit rating is more prone to BP neural network algorithm, while RBF neural network is applied less. Due to the confidentiality of privacy, personal credit data is difficult to be acquired. Domestic research is largely based on foreign data, with small volume and low data dimension, which is inconsistent with current domestic personal credit and loan business.To tackle the above problems, this thesis adopts the personal credit data of a domestic P2 P company named PPDAI to carry out research on personal credit rating. The data has numerous samples, large volume and high dimension. Therefore, lowering dimension operation is introduced before data analysis, and the variables of redundancy are eliminated. After that, sample data set before and after dimensionality reduction will be introduced to BP neural network and RBF neural network in order to calculating the results of comparative analysis.Results show that reducing dimension for high dimensional data can effectively lower computing time and increase the prediction efficiency of models. The advantages of strong RBF neural network like comprehensive approaching ability, simple structure and fast convergence rate will have higher prediction accuracy compared with BP neural network in personal credit rating.