The Research For "Case Mix" Classification Based On Rough Set And Classification And Regression Trees

With the fast development of date mining technology, the task of classification study of case mix, basing on rough set and classification and regression trees(CART), focuses on the deep-going on the study of case mix at home and abroad and the sharp rising expenses of medical treatment. The way of payment, which bases on case mix, has been paid attention to and studied extensively in the world in order to control and mitigate the expenses of medical treatment. Many countries have started to use the way. It is generally accepted as the most effective way, and it has practical meaning in many areas, such as hygienic policy, the management of medical treatment, and hygienic economy. Our country started to study case mix since 1990s and has developed the algorithm of AID(automatic interaction detection), case classification, and so on. These studies inquire into DRGS(diagnosis related group system) so they can promote the reform of medical treatment system and make full use of medical resource, which is meaningful in theory and practice. But because of our late study and the limitation of medical condition, we have no systematic study on case mix in our country now.This paper firstly introduces the present conditions of study at home and abroad, and presents the necessity of study in our country, then lists two ways of case mix which are in common used in our country: the algorithm of AID and case classification, at the same time, discusses the advantages and disadvantages of this theories. The advantages of AID are follow: First of all, it can select the target. At the same time, it can use the best method of partition to select a point automatically. And this point is the best point to this classification. The calculate is simple because it only split one target's variable in every time. The first way of the method of AID is mainly used to analyze numerical attribute. But in the study of case mix, the data is discrete, if we use the way of numerical attribute to analyze discrete attribute, the result is bad. The way of case classification is distinguishing the common case and the complex case. And then analyze the result. The advantages of case classification is consistent in the groups on consume the resource of sanitation. Thereby, it urge the quality's estimate of medical treatment reflecting the fact of medical treatment exactly. And making the manage of quality and the manage of the fee can be leading to the right. But the way of case classification has no such a standard which is extensively accepted, so the way exists the following instabilities: hospitals in different rank have different cognition on degrees of the state of an illness; the judgment to the state of an illness is affected by a doctor's technological level; the first diagnosis is perhaps affected by benefit because it is connected with the expenses; it is not so accurate to choose ICD-9 or the first three of ten coding to classify.Base on supplementing each other between decision tree theory and rough set theory , the paper discusses a new model of case mix, which combines rough set and CART. The model has the advantages of rough set. It can The attribute reduction of rough set can delete the redundant information in the knowledge successfully, and find the combination and rules hiding in the knowledge to help people to make correct and brief decision. The model has the advantages of decision tree also. It can classify data accurately, and has good abilities to study by itself and simple tree structure. Besides, The model can deal with the classification of numerical attribute and discrete attribute effectively, and shows an accurate and stable way for analysis. The model also can deal with more complex nonlinear data , interaction and the problems of lack of data. It is easier to grasp and explain the results. It includes three parts: rough set attribute reduction, CART and optimizing secondly procedures. The part of rough set reduce the number of attribute. The part of CART introduces the calculation of the development of CART, the pruning calculation of cost-complexity, and the calculation of test sample estimates for choosing the best tree. The part of optimizing secondly procedures is mainly about the best case mix, according to the characters of case mix and the reorganization to the best tree.We have used a system to realize this model. Apart from that, this classification system proves to be viable and reliable through the operation of data from Subordinate Hospital of Guilin Medical Institute. Through contrastive analysis with others, the model has many other advantages, such as the promotion in specification of classification, the full use of the respective advantages from rough set and CART, and so on.In the conclusion part, I make a conclusion to the whole paper, point out the existing problems, and have prospect of future work.