Study On The Decision Tree Classification Algorithm And Its Application Based On Rough Set Theory

In view of the fact that many decision tree classification algorithms are unable to deal with continuous attributes and missing values, this paper do some studies on decision tree classification algorithm based on rough set theory. The purpose of study is to build decision tree with higher accuracy rate and to enable decision tree classification algorithm to deal with continuous attributes and missing values effectively. In order to attain this purpose, this paper has done some studies on data preprocessing algorithms and decision tree building algorithms based on rough set theory. In view of the fact that discretization algorithms for incomplete decision tables are very few, an improved discretization algorithm based on attribute significance is proposed to discretize continuous attributes of incomplete decision tables. An improved ROUSTIDA is presented to solve some flaws of ROUSTIDA. A building decision tree algorithm based on rough set theory is presented in which attribute significance and approximate classification accuracy are used as splitting criterion of building decision trees.On the basis of the algorithms adopted and presented in this paper, a decision tree classification algorithm based on rough set theory is designed. The classification algorithm includes three steps:the first step is reading sample set and doing some preprocessing for sample set, the second step is building decision tree with the building decision tree algorithm based on rough set theory, and the three step is pruning decision tree using PEP. The classification algorithm can deal with continuous attributes and missing values, can build decision trees with higher accuracy rate as well. The experiment shows that the classification algorithm is of high performance.As for the application of algorithms, this paper designed a Classification System of Electronic Learning Product Selling. The main function of the classification system is classifying electronic learning product by sales quantity. The building decision tree algorithm presented in this paper is applied to the system, and the system prune the built decision tree with PEP. The system is implemented by SQL Server 2000 and VC++6.0. The result of system testing shows that the system can classify electronic learning product effectively by sales quantity, which to a certain degree, provide selling decision makers with valuable analytic method and decision support.