Survival Data Model Variable Selection

Survival analysis is a new branch of mathematical statistic since the 20th century 70years. A large number of practical issue is put forward by modern medicine, biology and other scientific research and so on.As the subject of the survival data for statistical analysis with great emphasis,it is used to estimate and predict problems in many fields, such as medicine, biology, actuarial science of insurance, engineering reliability, public health, economics, and demography and so on.Survival analysis is the subject to analysis one or more of the non-negative random variables,that is,the observed data according to its performance to conduct statistical inference. Statistical methods of survival analysis based on the content and theoretical development to date can be summarized as follows:basic data types, parameter model and maximum likelihood estimation, non-parametric method, semi-parametric model, regression analysis of censored data, hypothesis test and multivariate survival analysis,etc.For the COX model, after an unknown monotonic transformation,the model is equal to a linear function of covariates plus a random error, and the random error can be known or unknown, that is,the form of the linear transformation model is g(T)=-β'Z+ε.Firstly, the estimates of linear model g(·) is solved, and then Lasso method is used to compress the coefficient of survival model,and make certain coefficient become 0, then the use of AIC or BIC criterion is used to amputate the coefficient of 0 in order to determine the order of the model to achieve the gole of variable selection; and the improved fused lasso is using of Monte Carlo method to determine the adjustment of parameters. So that the model of the volatility is more smaller, relatively more stable. Finally, according the result of simulation data and real data to illustrate variable selection problems of the survival data. Then the problems need further researched is proposed.