Kernel-Based Residual Weighted Learning For Estimating Individualized Treatment

Personalized medicine has received increasing attention among statisticians,computer scientists,and clinical practitioners.Its main task in statistics is to estimate individual treatment rules.Many statisticians have proposed various innovative methods,including the RWL(Residual Weighted Learning)proposed by Zhou Xin in2017.RWL is a new and more optimal method based on OWL(Outcome Weighted Learning).Although the RWL is more widely used than the OWL,it can handle more complex treatment outcomes and more stable.However,according to the results of data analysis,RWL is not far behind the OWL in various measures,in some cases the OWL method still performs better and its calculation steps are complicated.In view of the above situation,the residual estimation method and calculation are improved in this paper.In terms of residual estimation,in order to get a better estimation,we use non-parametric estimation method to replace the linear estimation,and use weighted support vector machine to calculate ITRs,it is easier to operate and has fewer steps.Through simulation studies and real data analysis,the method we proposed performs better under various measures.