Application Research Of Random Forest Method Based On Partial Least Squares For TCM

The TCM data have the characteristics of multiple components,multiple effect,multiple targets,nonlinear and so on,which determines the TCM data present an nonlinear relationship with multiple independent variables and dependent variables.Meanwhile,because of the TCM experimental data raising daily and the limit of the objective causes,the utilization of the data often stays in the basic analysis stage.Therefore,it is indispensable to study a deeply analysis method with multiple and nonlinear suiting the features of the TCM data.So,the paper attempts to combine the Random Forest Algorithm with the Partial least squares,and then apply that to the TCM experimental data,finally provide technical support for excavating the Chinese medicine information.The primary work of this paper is as follows:Firstly,because when the built-in T2 elliptic map of the PLS has restricts in analyzing the specific point in the high dimensional space,the paper proposes to recognize the specific point of the TCM data by using the SBM model,and then through comprehensive evaluation and screening out the effective data to implement the PLS regression,and after that the paper uses the Dachengqi Decoction experiment about the experimental intestinal obstruction changing entire party dose and the classic tool wear experiment to verify that.Finally the experimental results show that the average relative error reflects a certain extent decline.Secondly,on the basis of the Model Tree and the PLS,the multivariate and regression method of the leaf node of the Classical Regression Tree is improved to be the PLS or the mean value to decide and choose automatically according to the sample information,namely the Partial Model Tree.Then the Ensemble Learning is applied to it,and an algorithm called the Random Forests of Partial Least Squares Regression.Afterwards it is verified on the monarch drug asthma experiment and monarch drug cough experiment of the Maxingshigan Decoction and the five UCI sample data sets of the Ensemble Learning.Finally the experimental results show that the explanation extent of the model gains certain improvement and it provides that this algorithm has certain feasibility and validity.Thirdly,according to the variable project importance analysis of the PLS and combining with the Hill equation with four parameters,the paper proposes a dose threshold evaluation algorithm based on the variable project importance analysis.And then the paper tests it on the experimental data of the Dachengqi Decoction about the experimental intestinal obstruction changing entire party dose and the experimental data of the Gegenqinlian Decoction about the ulcerative colitis changing entire party dose?Finally,the paper obtains that the dose of the intestinal obstruction clinical patients treated by the Dachengqi Decoction should not be recommended the low dose,and the dose of the ulcerative colitis patients treated by the Gegenqinlian Decoction should be advised the middle dose.Fourthly,basing on the above research achievements and referring to the analysis requirements of the TCM data,the paper designs and develops a TCM data analysis system based on the PLS-RF ultimately using JAVA.