Research On Micro-credit Mining Technology For Weakly Usable Data

In order to obtain the funds needed for development,small and micro enterprises usually need bank loans.With the advancement of computer technology,the traditional offline loan is gradually replaced by the online loan.User credit rating and risk prediction have always been the most critical links in the online loan process,but there still exist some problems.Firstly,the quality of user data is poor and the problem of weak availability is ubiquitous,which can't guarantee the authenticity of the data.Secondly,lacking no scientific rating system for user credit rating,which cannot reflect the objectivity of the evaluation results.Thirdly,the prediction process of user's loan risk cannot effectively utilize the qualitative credit index,which cannot ensure the reliability of the prediction results.To obtain more objective and reliable user credit rating evaluation results and user loan risk prediction results,in this paper,the mixed interpolation method is used to repair user data which is weakly available.That is to say,different processing methods are adopted for data with different missing rates.Due to the large amount of information in user data,it is not only inefficient but also not objective if the users' credit ratings are divided one by one.This paper proposes to evaluate the basic credit rating of users by taking advantage of the Multi-task learning ideas.This method utilizes the Shared structure relationship among multiple credit ratings to construct a training network to implement the credit ratings of users.At the same time,combined with the initial credit rating of users,the fuzzy Logistic regression method is used to predict the default risk of usersThe main contributions of this paper are as follows.First of all,in this paper,a mixed interpolation algorithm is proposed to complement the missing value of user credit data.Secondly,it is proposed to use a Multi-task learning model to evaluate user credit rating.Thirdly,it is proposed to use the Likert scale to quantify the credit index,use the triangular fuzzy number to quantify the credit index,and construct a fuzzy Logistic regression model to predict the probability of default by the user Finally,a small and micro-credit system is designed and implemented,and comprehensive experiments demonstrate the effectiveness and feasibility of the proposed method.