| An important branch of model research in survival analysis is the AFT model.The AFT model is usually used in the study of censored data because of its simple form and easy to interpret.But another more common model of survival analysis is Cox model,due to the abuse of the Cox model,the AFT model has made little progress in the selection of variables.In order to expand the research on the related properties of the AFT model,the relevant properties of the AFT model under the LASSO and MCP penalty methods will be inferred: Based on the MCP penalty method’s Oracle property,the penalty items in the AFT model with censored data will be adjusted,and then the MCP method in AFT model can show better properties such as variable selection consistency and asymptotic behavior.The coordinate descent(CD)algorithm is used in the process to accelerate the convergence,and a fast iteration is performed in the penalty function to find the local minimum.The simulation shows that the MCP method has better properties than LASSO,it can select more accurate and simple variables,and also has less mean square error,and the MCP method has no difference after adjustment,but the LASSO method is more inclined to put the coefficient down to 0.Generally,the MCP method has a better nature.The study enriched the selection of variables in survival analysis with censored data.The case study analyzed the financial status of real estate listed companies,and used the theory to introduce the excellent properties of the AFT model under the MCP method.The 216 variables are finally compressed to 16 variables,The MCP method has a better fitting effect and expands the survival analysis in financial risks. |
