Interactive Visual Classification Based On Parallel Coordinates

With the continuous development of data mining, visual data mining is rising gradually. Thanks to visualization technology, one can observe the distribution of the data, understand the process and results of data mining, and even directly involved in the data mining process.This paper brings up a rather simple visual classification algorithm-Scanned-branch tree algorithm (SBT). SBT scans each dimension of the data set, extracts decision branches (rules) directly from the properties, and finally builds a classification model (SBTree), which is a single-branch multi-leafy tree in appearance and an ordered sequence of rules in essence.Experiments prove the SBT is feasible and can be used in interactive visual classification by combining it to parallel coordinates. User can find and manually extract decision branches on parallel coordinates. By such an interactive way to establish classification model, user can deepen the understanding of the training set, and will be more confident of the classification model. Most importantly, field knowledge and the initiative of human can be played perfectly.As a new interactive visual classification method, this paper will expound it by an actual example, and also will discuss its ability to deal with all kinds of data sets. Thus, this paper try to treat the method as a system and it can handle any cases that can be dealt with by traditional decision tree.