Combination Of Deep Belief Network And Random Forest For Personal Credit Assessment

Nowadays,the development of Internet,big data,artificial intelligence and real economy has already complement each other.Building a bridge between information technology and financial economy is a new grasp and new motive force to cultivate China's economic growth.In the daily economic activities,the material development is constantly rich,the standard of the good life that the individual yearns for is also constantly improved,and the development scale of the personal credit business is constantly expanding,which makes the work in this field more difficult.Therefore,how to explore more scientific and efficient credit risk assessment methods has become one of the hot spots in the direction of financial science and technology.As financial activities permeate everyone's daily life,massive data make it more difficult to describe personal credit information,and at the same time,different characteristics contribute different degrees to credit evaluation.Therefore,how to learn more accurately the decisive characteristics that affect credit is the key to research.In machine learning,the bionic neural network of human nervous system and cognitive mode can efficiently synthesize local information in massive data.This makes the process of feature extraction effectively reduce the influence of human subjectivity compared with traditional methods.Deep Belief network based on restricted Boltzmann machine network,deep trust network can deeply mine the relationship between features.Random forest is widely used as an ensemble learning method,which can make the base classifiers learn from each other's strong points,and at the same time,can effectively avoid the over-fitting of a single decision tree.Therefore,this paper combines confidence network with random forest to study the personal credit evaluation.The main contents of this paper include the following three parts:A personal credit assessment model based on random forest is constructed.Random forest is a common integration method,which has achieved good results in dealing with classification and other problems.By German credit data and X commercial bank of China credit data classification performance contrast experiment on two data sets can be found that using the method of random forests the credit data are important feature extraction and outlier removal can effectively improve the classification performance of the models,X in China commercial bank credit data accuracy reached 98.23%,However,the classification performance of German credit data needs to be improved.A personal credit assessment system based on confidence network is constructed.With the rise of the concept of artificial intelligence,machine learning has evolved in the direction of being more human-like.This makes machine learning a trendy way of processing information,mimicking the learning of biological nervous systems.In this paper,a personal credit evaluation index system is established by using the confidence network.Based on the German credit data set,a total of 17 important funds were extracted,which were classified into 12 categories of attributes,including credit history,current checking account status,loan purpose and other important attributes.Based on the credit data set of B ank X of China,a total of 15 important items were extracted,which were classified into 9 categories of attributes,covering important attributes such as credit history,length of service,occupation and housing.A portfolio model of personal credit evaluation combining conviction network and random forest is constructed.Considering that confidence network can comprehensively learn credit characteristics without supervision,a combination model of individual credit evaluation is constructed by combining confidence network with random forest.By comparing with the three classical methods of random forest,association rule-based classification and classification decision tree,the combination model has the highest accuracy,best generalization performance and strongest robustness for credit evaluation in both the German credit data set and the Chinese commercial bank data set.