Scoring card fusion method and device, computer equipment and storage medium

Abstract

The invention discloses a score card fusion method and device, computer equipment and a storage medium, and belongs to the technical field of risk control. The method comprises the steps that an expert score card and a machine learning score card are acquired, and feature variables used by the expert score card are different from feature variables used by the machine learning score card; the expert score cards and the machine learning score cards are respectively fused on the plurality of weight combinations to obtain a plurality of fused score cards; and the plurality of fusion score cards are verified by using the verification sample, and the optimal fusion score card is screened out. Compared with the prior art that variables selected by experts are directly added into the machine learning model, the process is repeated until the variables meeting the model conditions are selected, the problem that the repeated workload is large can be avoided, the problem of sample data deviation can be effectively avoided, and the model prediction capacity is improved.

Description

technical field [0001] The present invention relates to the technical field of risk control, in particular to a scoring card fusion method, device, computer equipment and storage medium. Background technique [0002] In the field of risk control technology, taking credit risk control in the financial field as an example, financial institutions usually need to build a risk control model to evaluate the credit risk of business objects. When evaluating credit risk for business objects, it can be used in the risk control model The scorecard performs credit scoring on business objects. The higher the score, the lower the corresponding credit risk, and vice versa. [0003] Scorecards can be divided into expert scorecards and machine scorecards. When a product has accumulated a large amount of sample data, it can be modeled directly using machine learning. However, when the product does not have enough data for modeling, or the new product has no data accumulation (unsupervised le...

AI Technical Summary

Problems solved by technology

Taking logistic regression as an example, directly adding a new variable in the regression formula may lead to a decrease in the significance of a variable, which is directly reflected in the increase of the variable p value, exceeding the threshold, then the added variable must be replaced until the selected variable is selected. In order to ensure that the model indicators meet the requirements, this method requires a large amount of repeated work, and the variables selected by experts do not necessarily have strong predictive power on the data selected for modeling (usually expressed as IV values low), there is a certain contradiction in itself to measure the variable of expert experience with the variable requirements of the machine learning model

Images