Reseaerch On Decision Tree And Application In Coronary Heart Disease Treatment

In recent years,Coronary Heart Disease has brought huge health risks and economic burden to human with the characteristics of high incidence and mortality.The development of computer technology provides technology basis to explore rules and medicine-used knowledge which is contained in Traditional Chinese Medicine(TCM)diagnosis and treatment.How to mine rules and knowledge effectively from existed data and how to use the rules and knowledge to provide the aided decision for the diagnosis and treatment of coronary heart disease are two contents mainly considered in this paper.The result got by Decision Tree is intuitive and easy to understand.To reflect the relationship between the symptoms of Coronary Heart Disease and syndrome result intuitively,multi-labeled and multi-valued Decision Tree is studied.However,the algorithm which is used to deal with the data exists some questions.Aiming at these questions,this paper makes some modifications,the experiment is designed to prove the effectiveness of these modifications.The specific research are as follows:First,on the process of choosing the splitting property,the original multi-valued and multi-labeled algorithm ignores these data which current value is empty.The data of Coronary Heart Disease has the characteristic of many null-values,so the original algorithm lose many data and has a low classification accuracy.Aiming at this problem,this paper makes some modifications about the selection of the algorithm's property,introduces the judge of the null-values,and puts the data whose current property is null as the new child node.Thus,the data is not lost.Second,if the data has the property with more null-values,the above-mentioned algorithm will produce over-fitting effect easily,so the decision tree has a large scale and the model has a low classification accuracy.Aiming at this problem,this paper introduces the threshold of the null-value number before building the Decision Tree,preprocess the data set according to the threshold,and excludes the data with more null-values.Thus,the problem of the classification's accuracy rate decreasing fast can be handled.Third,in the stage of the evaluation of division effect,a new formula of similarity computation is proposed in this paper.It will make the similarity between two label sets compute more reasonable,the parameter in the formula can reflect the characteristic of the sets,the algorithm can be adjusted automatically according to the characteristic.Fourth,on the aspect of actual application,a system for aided diagnosis and treatment is designed,the system adopts this modified algorithm.The system will forecast the syndrome results according to the selected symptoms,the forecasted result will be got and offered as references to doctors.In addition,the training data set of the classified models can be dynamically increased,and we can extend the specific mining sub-modules for different purposes based on this system.