Single-terminal multi-user risk control modeling method based on Bayesian network model

Abstract

The present invention provides a single-terminal multi-user risk control modeling method based on the Bayesian network model, which is characterized in that it includes the following steps: collecting the information of all accounts logged into the current terminal device during the historical time period when no risk event occurred Historical behavior data, based on historical behavior data, the organizational structure relationship between accounts and the order in which each account logs into the current terminal device uses the decision tree model to calculate the intimate relationship between accounts and accounts; establish a complex network model; when complex networks When a high-risk scenario occurs at a certain node in the model, the complex network model is regarded as a Bayesian network, so as to determine the risk of each other account in real time.

Description

technical field [0001] The invention relates to a method for controlling the risk of multiple accounts logging into the same terminal device. Background technique [0002] In enterprise applications, since many devices are fixed assets purchased by the enterprise, in most cases the device coupling cannot represent the association of risks. For example, in a bank, several tellers share one device. For the above situation, risk monitoring is usually carried out by using risk control models based on logistic regression, decision tree and artificial neural network, but the above risk control models are all batch execution and batch judgment models, and cannot achieve real-time monitoring. [0003] In fact, even if it is technically possible to use the above-mentioned risk control model for real-time monitoring, in practical applications, it is impossible to use the risk control model based on logistic regression, decision tree and artificial neural network to effectively contro...

AI Technical Summary

Problems solved by technology

For the above situation, risk monitoring is usually carried out by using risk control models based on logistic regression, decision tree and artificial neural network, but the above risk control models are all batch execution and batch judgment models, and cannot achieve real-time monitoring

[0003] In fact, even if it is technically possible to use the above-mentioned risk control model for real-time monitoring, in practical applications, it is impossible to use the risk control model based on logistic regression, decision tree and artificial neural network to effectively control the risk of single-terminal multi-user scenarios

If the parameters of the model are relatively stable, that is, it is necessary to change the estimated value of one or several parameters, a large amount of new data may be required to affect the results of the model, and the emerging risks cannot be judged in real time, so that the risks cannot be prevented

If the parameters of the model are relatively unstable, that is, very little data can change the estimated value of one or several parameters of the model, it is easy to cause underfitting of the model

[0004] In summary, the risk control models based on traditional logistic regression, decision trees and artificial neural networks lack "elasticity" in real-time monitoring scenarios, that is, they cannot respond to real-time changes in data efficiently and timely.

Images