Model prediction method and related equipment for predicting the probability of overdue repayment by users

Abstract

The invention discloses a model prediction method, device, equipment and storage medium. The model prediction method includes: obtaining prediction models of N stages, the prediction models of different stages are obtained by training based on different data sources, and N is greater than An integer of 1; the sample is predicted according to the prediction model of each stage; and if the sample is predicted according to the prediction model of the kth stage, the prediction result of the sample according to the prediction model of the kth stage will be used from participating in this stage. The predicted samples are selected to participate in the prediction of the next stage, and k is any positive integer less than N. According to the embodiment of the present invention, the amount of data for calling external data sources can be reduced, and the prediction effect of the model can also be guaranteed.

Description

technical field [0001] The invention belongs to the field of computers, and in particular relates to a model prediction method and related equipment for predicting the probability of a user's overdue repayment. Background technique [0002] The current machine learning algorithm can accept thousand-dimensional feature input. These features come from different external data sources. Every time the prediction model is used for sample prediction, it is necessary to call all the data sources used at the same time, and the interface calls of different data sources. There is not only the time cost of calling, but also the corresponding data cost. [0003] In response to the above problems, the traditional approach is to reduce the original thousand-dimensional features to an acceptable range based on feature screening technology, which can ensure that the amount of data from external data sources is called as little as possible, but the cost is the prediction effect of the model ...

AI Technical Summary

Problems solved by technology

[0003] In response to the above problems, the traditional approach is to reduce the original thousand-dimensional features to an acceptable range based on feature screening technology, which can ensure that the amount of data from external data sources is called as little as possible, but the cost is the prediction effect of the model The decline is more obvious

Images