Cox Optimization Algorithm Based On Gradient Boosting Neural Networks

In survival analysis,Cox proportional hazards model(Cox model)is widely used by researchers because of its simple calculation and easy understanding.However,Cox model has many limitations in practice,and the existing optimization models are always unsatisfactory when used.For example,the penalized Cox optimization models are still not suitable for situations where the logarithmic risk function and covariates exhibit a nonlinear relationship,Cox optimization methods based on neural networks are more used in low-dimensional scenarios,and Cox optimization models based on gradient lifting algorithm framework require too many parameters to be adjusted.Therefore,it is of great practical significance to study how to optimize Cox model.Based on Cox model and Gradient Boosting Neural Networks(GrowNet)algorithm,this thesis proposes a Cox optimized survival analysis model-GrowSurv.This new algorithm can effectively depict the complex nonlinear relationship between logarithmic risk function and covariates,solving the limitation that the logarithmic risk function in Cox models must be a linear combination of covariates,and has relatively wider application.When solving the objective function,the GrowSurv model is less likely to fall into the local optimal value,and the number of parameters to be adjusted is less.Compared with the common survival models Cox,Lasso Cox,XGBoost Cox,Deep Surv and random survival forest,experiments on the simulation data set and the real data set show that the prediction performance of GrowSurv model is good.Especially on high-dimensional data,GrowSurv model is still stable.Finally,this thesis uses the new algorithm to build the prognosis model of heart failure.The experimental results fully prove that GrowSurv model has good risk stratification ability and prediction performance.