Research On Internet Personal Consumption Loan Pricing Based On Clustering Algorithm

There is no doubt that the issue of personal Internet loan pricing is a key issue for many financial systems.It concerns the interests of both parties to the loan.It is a new issue in the new Internet-based business model.Related loan pricing technologies are also continuously making progress.Advanced loan pricing models have also sprung up,and related research has made a lot of progress.Loan pricing is based on the consideration of loan risks and maturities by commercial banks based on their own capital costs and profit targets,and after combining the supply and demand of funds in the lending market and the user's loan situation For research questions that have only been made,in order to target different user groups,according to their personal Internet consumption loan situation and user's personal situation,a model is designed to evaluate the risk level of different loan users,and combined with other optional The parameters jointly design a reasonable classification model,and design personalized loan pricing services for users to maximize the economic benefits of both parties,which has great practical significance.Doing research on personal Internet loan pricing is an important condition for Internet loans to gain a comparative advantage among the major Internet banks.This study is to some extent conducive to increasing capital dispatch and resource allocation in the additional debt industry,and it is also conducive to lending platforms.To further develop the financial market,the traditional model or research method is not seriously out.The traditional model or research method of loan is mainly based on the loan interest rate,personal income and risk control,market system and other factors to analyze loan pricing.This research was conducted using machine learning and big data technology for modeling,which resulted in a multi-angle analysis of the main factors affecting debt pricing.An A-Kmeans clustering model and hybrid structure suitable for this application were proposed for VGG19 connection 10 Layer DenseNet structure and 6-layer ResNet model to solve user risk grading problem and loan pricing problem respectively.The A-Kmeans clustering model consists of the data dimension reduction and Kmeans clustering stages of AutoEncoder.The data dimension reduction operation is used to be able to Compress the main data characteristics,taking into account the Kmeans algorithm In clustering in low-dimensional space,the effect can only be clustered by Kmeans algorithm by nearly 4 subdivisions.At the same time,in order to achieve the fit of the model to the data and improve the parameter sharing,self-encoding in AutoEncoder The encoder also uses a 10-layer DenseNet structure and a 6-layer ResNet structure as encoders and decoders,so that parameter sharing can be performed with this part of the data classification model,and the model sharing in the same set of data is guaranteed to the greatest extent.In the model,the structure of VGG19 is used to easily extract data features.The DenseNet connection mechanism is used to reduce the complexity of the model,reduce the amount of parameters,improve the sharing of parameters,and reduce the disappearance of the gradient of the model.ResNet is used to be able to Standardize important data characteristics,so the output of the model is closer to the number of classifications.Based on this,what are the main factors related to loan pricing and risk assessment,and provide personalized pricing services for customers' loan business,and On this basis,the effect of each fundamental parameter is visualized to clearly reflect the impact of wind.The mAP index is used as the accuracy index of the evaluation model.The experimental results show that the AutoEncoder model used in the data dimension reduction problem achieves an accuracy of 94.5%.The earlier PCA model and LDA model are nearly 5 subdivisions higher.The Kmeans model used in the risk rating problem achieved 88.3% accuracy.The earlier Pmeans and DBSCAN models were nearly 3 subdivisions higher.The VGG19-10D-6R model used in the loan classification problem achieved 87.9% accuracy.The accuracy of other earlier feature extractors is comparable,but the amount of parameters is nearly 10 MB smaller,and the test takes less than 10 ms.At the end of the experiment,the design of the number of layers of the DenseNet structure and the ResNet structure were compared,the number of loan pricing classifications was compared,and the number of classifications of the risk level model and the loan pricing model were comprehensively compared.It was found that a 10-layer DenseNet structure and a 6-layer ResNet structure were used.As a classifier and encoder,the effect is nearly 1 higher than the accuracy of other structures.The effect of using 2 loan classification numbers is higher than the accuracy of other structures by 2 subdivisions.The amount of parameters is nearly 4MB,and 5 risk levels are used.As the basis of the credit classification task,the class is nearly 1 variable higher than other structures,and the amount of parameters is nearly 5MB smaller.And finally found that the factors that affect the risk rating are mainly whether the loan is overdue and the status of the account under the name.The factors that affect the classification of loan pricing are mainly through the construction of a scientific feedback system to obtain customer feedback in real time and continuously optimize us.Model,the transfer of generalization capabilities of the network structure,and the model has excellent performance.In the future,the model can be deployed on a cloud server,collect real-time data of the loaned customers through the APP platform,and conduct reasonable loan pricing,and provide timely feedback to the pricing results.Model new data information to upgrade the model.