A CR Tree Based Data Ming Visualization Framework

With the development of data mining technology,data mining algorithms are be-coming more and more complicated.It makes the demand for data mining visualization more and more intense.There are three motivations to build data mining visualization system:1.To help beginners and users understand the working principle of data min-ing algorithm;2.To explain the process of data mining,intermediate results and other details;3.Auxiliary data mining technology research.There are many literatures proposed visualization systems for naive bayesian,de-cision tree,neural network and other data mining models.However,these literatures are limited to the UI design,did not propose a more general data mining visualization solution.The research of this thesis focuses on data representation,storage structure and UI design for decision tree and neural network visualization system.This the-sis proposes a CR tree-based data mining visualization(CR-DMVF)to help user build visualization system easily.The CR-DMVF focuses on two issues.The first is how data mining data is rep-resented and stored.In this paper,the intermediate data involved in decision tree and the neural network training process are represented by(time,node,edge,attribute)and CR tree is proposed as index structure for visualization system.The second is how to design the UI for data mining visualization system.The framework defines navigation view,model view,data view and timeline.The most important contributions of this work are the following:1.This thesis presents a CR tree-based data mining visualization framework(CR-DMVF).CR-DMVF divides the data mining visualization system into four lay-ers:UI,visualization,data management and data mining model.The UI layer defines navigation view,model view,data view and timeline.In the visualiza-tion layer,the thesis introduces the initialization of the navigation view.The data management layer consists of node index,intermediate data index,and data filter.2.This thesis proposes that the intermediate data of the decision tree and the neural network algorithm can be represent by a quaternion consisting of time,nodes,edge and dataset attribute.3.Faced with the low search efficiency because R tree often produce slender rectan-gles,the thesis propoeses CR tree.CR tree improves the node splitting algorithm.When the node need to be splited,CR tree will calculate radio of edge of the rect-angle and the coordinate space rectangle and select the edge with the largest ratio as splitting edge to suppress elongated rectangle generated.Then,the standard points will be calculated and the sub-nodes will be divided into two groups ac-cording to the distance between the centroid of the node and the standard points.Experiments show that CR tree greatly improves the performance of insertion,deletion,point search and range search.