Variable Selection For Continuation Ratio Model And Its Application

Ordinal categorical data is very common in medicine and histology.Along with the development of DNA technology,more and more medical research set out to explore the relationship between diseases and gene expression of patients,but gene expression data is usually ultra-high dimensional.Traditional statistical methods that can solve the ordinal categorical problem all require the variable dimension is smaller than the sample size,therefore they are not useful in high dimensional case.In order to solve the high dimensional ordinal categorical problem,we based on general continuation ratio model and consider the group effect of coefficients of the same variable under different categories,add 1-norm MCP penalty function into model and propose 1-norm MCP penalized continuation ratio model.We first do simulation to compare the performance of the 1-norm MCP penalized continuation ratio model with L1 penalized constrained continuation ratio model and L1 penalized continuation ratio model.Simulation results show that the performance on variable selection and classification of 1-norm MCP penalized continuation ratio model is better than the others,this means considering the group effect of the coefficients,take the coefficients of the same variable in different categories as a group,punish and estimate the coefficients of the whole group at the same time can effectively improve the performance of the modelIn addition,we apply the 1-norm MCP penalized continuation ratio model into credit risk data,prostate cancer classification gene expression data and bowel disease diagnosis gene expression data,the analysis results show that the model can effectively select the important variables that associated with the response and its classification accuracy is high.