Method and device for updating tree model in equivalent interval, computer equipment and medium

Abstract

The invention relates to a method and device for updating a tree model in an equivalent interval, computer equipment and a medium. The method comprises the following steps: acquiring a test set, wherein the test set comprises customer feature data of each customer in a current customer group; obtaining an equivalent interval table corresponding to the target tree model; calculating a value of a customer group displacement index corresponding to the test set according to the equivalent interval table; verifying whether the customer group displacement index is related to the prediction accuracy of the target tree model or not, if so, monitoring the value of the customer group displacement index, and updating the target tree model when it is monitored that the value of the customer group displacement index exceeds a preset customer group displacement threshold value. By adopting the method, a plurality of customer feature changes can be converted into a single simple measurement value to judge whether the customer group feature has displacement related to the model prediction accuracy, so that the tree model can be updated in time, and the prediction accuracy of the tree model is ensured.

Description

technical field [0001] The present application relates to the technical field of model updating, in particular to a method, device, computer equipment and medium for updating tree models in equivalent intervals. Background technique [0002] In the field of financial anti-fraud, machine learning models based on decision trees (ie, tree models) are often used to predict fraud risks for customers. However, as time goes by, the prediction accuracy of any tree model for new customer groups will inevitably attenuate. There are two main reasons for this attenuation: the first is that the tree model has appeared in the distribution of customer group characteristics The second type of new customer groups I have seen is that the distribution of customer group characteristics is stable, but the correspondence between customer group characteristics and labels has changed. If it is the second reason, it must be known after the customer's label is observed, and the label is usually obta...

AI Technical Summary

Problems solved by technology

Among them, method one can only measure the change of a single feature, and cannot simply and intuitively measure the displacement of the overall customer group, and ignores the influence of features on the prediction score (the change of features does not necessarily lead to the change of the prediction score); method two is the same as the prediction score However, there is still the possibility that the overall customer group displacement and the predicted score are not related, and there are usually more than one important features of the model, and the calculation is cumbersome; Method 3 can detect changes in the predicted score, but the change in the predicted score does not necessarily mean that new customers The emergence of a certain group may be the result of a large number of identified customer groups (such as high-risk customer groups) appearing in a certain period of time, so the change of the prediction score cannot fully and necessarily predict the accuracy of the model. and the emergence of new customers

Images