Survival analysis risk function prediction method based on gradient survival promotion tree

Abstract

The invention relates to a survival analysis risk function prediction method based on a gradient survival promotion tree. The method comprises the following steps: S1, obtaining survival analysis sample data, and carrying out the preprocessing; s2, constructing a gradient boosting tree prediction model based on a greedy segmentation algorithm; s3, training the gradient boosting tree prediction model by using the training data; and S4, inputting the preprocessed survival analysis sample data into a gradient boosting tree prediction model, and outputting a risk function obtained by prediction. Compared with the prior art, the method has the advantages of accuracy, generalization performance, interpretability and the like.

Description

technical field [0001] The invention relates to the field of survival analysis, in particular to a survival analysis risk function prediction method based on a gradient survival promotion tree. Background technique [0002] Survival analysis, commonly used in actuarial and biomedical fields, is currently a hot model in the field of credit risk statistics. Different from traditional regression problems, the research goal of survival analysis is the probability of an event occurring at a specific time point, and then estimate the user's survival curve over time, rather than just predicting a target variable. Traditional survival analysis methods generally make certain assumptions about the risk function of individual users, and set the model parameters to have a linear relationship with individual covariates. The predictive performance of the model is closely related to the accuracy of the assumptions. Once the assumptions are not accurate, the model predictive power will be ...

AI Technical Summary

Problems solved by technology

Although the commonly used GBDT and XGBoost can fit the user's risk at a specific time point in survival analysis, when fitting the user's risk at multiple time nodes at the same time, the loss function used by these methods is not accurate enough and cannot Obtain satisfactory effect in the field of the present invention

[0007] The survival analysis model based on the neural network requires massive data and constant adjustment of parameters. Although the prediction performance of the model is not bad, because the prediction model is a black box, the interpretability is not enough, so the field of survival analysis generally does not use depth. learn related technologies

Images