Risk prediction method based on sorting and device thereof, equipment and medium

Abstract

The invention provides a risk prediction method based on sorting and a device thereof, electronic equipment and a medium, aiming at overcoming the defects of data potential safety hazards and the like caused by inaccurate prediction due to business data sample deviation of an existing business risk prediction model, and aiming at solving the problem of how to establish a training data set by introducing a user sample without credit performance data. The model is constructed based on training data sorting, and then the constructed model is used for target user risk prediction, so that the model prediction precision is effectively improved under the condition of sample deviation or sample data missing and the like, and the prediction evaluation of the model is correctly and accurately completed; the technical problem that consideration factors of model risk assessment are more comprehensive is solved, and risk loss caused by data potential safety hazards is avoided.

Description

technical field [0001] The present invention relates to the field of Internet information processing, in particular to a sorting-based risk prediction method, device, electronic equipment and computer-readable medium. Background technique [0002] With the rapid development of Internet finance and the change of people's consumption concept, advance consumption by way of credit is becoming more and more accepted by people. Due to the credit financial data transmitted and processed through the network, the requirements for information security and the credibility of related data will be higher. Taking the credit scenario of Internet finance business as an example, whether users applying for loans provide or upgrade their credit data, restrict the use of corresponding data, etc., need to consider the possibility of their overdue. In this way, it is necessary to establish a credit scoring model based on their various behavioral information and demographic characteristics as inp...

AI Technical Summary

Problems solved by technology

A condition for the successful application of machine learning algorithms is that the training samples and test samples have the same distribution, but there are certain defects in the credit scoring model construction model data set, which cannot fully utilize the information of all data

When building a machine learning credit scoring model, sufficient labeled data is required. The more the data can reflect the information model of all samples, the higher the accuracy of the model is. However, the data obtained by organizations that provide data on the business platform are often incomplete data that are not randomly missing. For example, in the financial business scenario, the lending institution has the relevant characteristic attribute information of all applicants, as well as the loan record and repayment record information of the users who have passed the application and have financial performance data. The credit records of users who have no financial performance and those who have no financial performance data such as those who have rejected loan applications are missing. These records are the best labels for building models, which will lead to inaccuracies in actual modeling. Sample non-random deviation occurs, which cannot accurately reflect the full sample, which leads to biased parameter estimation during modeling, thereby affecting the judgment of real risks, making accurate judgments impossible, and easily causing credit data to be insecure, credible, and incorrect. problems, which in turn lead to corresponding hidden dangers of data security in various business scenarios, such as economic losses of lending institutions due to financial default in financial business

Therefore, the existing user risk assessment / prediction technology has sample bias, which affects the accuracy of the assessment and cannot really meet the requirements of user default risk assessment

Images