Combination Of Deep Autoencoder And Support Vector Machine For Personal Credit Assessment

Since the reform and opening up,Chinese socialist market economy system has become more and more perfect.In order to expand the domestic market demand,personal credit lending platforms such as bank credit loans and ant flower buds have been constantly rolling out,and personal credit consumption behaviors are increasing.Personal credit assessment has received the attention of the government and enterprises,and has become one of the research hotspots of the financial industry,especially the banking industry.The depth autoencoder can extract the depth features of the data while the kernel support vector machine has high classification accuracy when processing "small samples,big data".This paper combines deep autoencoder and kernel support vector machine technology to evaluate the personal credit rating.The main contents of the thesis include the following parts:Constructing a personal credit evaluation model based on diffusion kernel support vector machine.The diffusion kernel is a kernel function constructed by using the adjacency relationship of the graph,and achieves good effects in dealing w:ith image classification and the like.In this paper,the kernel is introduced into the support vector machine,and a personal credit evaluation model based on the diffusion kernel support vector machine is proposed.Experiments show that the personal credit evaluation ability of the diffusion kernel is higher than that of the polynomial kernel,the Sigmoid kernel and the Gauss radial kernel;in Germany,credit history,current checking account status,other debtors,loan purposes,deposits,property,housing,working credit history,working hours,housing and occupation are important attributes for evaluating personal credit;while in China,credit history,service age,occupation a nd housing are vital.Building up a personal credit assessment model using integration of multi-kernel support vector machines.According to the different data types of the index attributes,four kinds of kernel functions,such as histogram cross kernels,diffusion kernels,generalized cosine distance kernels and Jaccard distance kernels are selected.Based on prediction results of these four kernel support vector machines,the weighted voting method is used to make final decision fusion.Experiments show that the method of integration of four kinds of kernel support vector machines can improve the evaluation performance of the model;the important indicators are refined by the model.In Germany,existing checking account status,other debtors,deposits,houses,jobs are the important attributes of credit evaluation;in China,credit history,income and occupation are important attributes for evaluating personal credit.Establishing combination of deep autoencoder and kernel support vector machine for personal credit evaluation.Deep autoencoder is introduced to encounter redundant information in processing personal credit data.Using the advantages of deep autoencoder to extract depth features and the high classification accuracy of fusion multi-kernel support vector machines,a personal credit evaluation combination model is proposed.Experiments show that the evaluation ability of the combination is higher than that of the multi-kernel support vector machine model;The important indicators has further been refined by the model.In Germany,estate is a deep attribute for evaluating personal credit;in China,social identity and credit record are deep attributes for evaluating personal credit.