Study On Diagnosis Related Groups Classification Based On Decision Tree

With the increasing volume of medical institutions,the hospital has entered the era of big data.Medical institutions can improve their management,diagnosis and clinical research by a reasonable and efficient data mining analysis tools,which is effectively,timely and accurate processing,In hospital management,the application of medical big data mainly focuses on the related application of DRGs.DRGs is the abbreviation of Diagnosis Related Groups,It's also called diagnostic groups in China.DRGs is based on the medical records,according to the clinical similarity(such as illness,treatment types and age,etc.),and resource consumption of similarity(such as length of hospital stay and cost),the number of all disease is divided into a number of disease-groups.Decision tree classification has been successfully applied to classification problems in many fields,but it is not widely used in DRGs.The decision tree classification is flexible,intuitive,clear,reliable and efficient.It has a great advantage in the DRGs classification prediction.In this paper,I make a research in the decision tree techniques on CART,CHAID,C4.5 and Random Forest algorithm.The DRGs data is come from a top-three-level hospital,and it's fed back by Guangdong Health and Family Planning Commission.I make an experimentation on DRG classification by using SPSS and Weka.I find that:(1)There are a lot of attribute values of prediction variable and target variable in the DRG classification.The prediction variables are decomposed before classification forecasting.And then,it's also necessary on one MDC DRGs classification prediction to another.(2)According to the research,the decision tree algorithm has different performance in the DRG grouping of MDC.And the result shows that the predictive accuracy of DRGs based on decision tree is very good,which the lowest group's accuracy rate is 96.87%.(3)The DRG grouping of each decision tree algorithm is compared in detail.And I got all the DRG classification rules from the best decision tree algorithm of each MDC.