The class imbalance problem is an important issue in classification of Data mining. Among the proposed approaches, some of them modify the class distribution of the original data which would worsen the computational burden or might throw away some userful information; some are limited to specific dataset or only applicable to the dataset with numeric attribute; some would take a lot of training time due to the natural property of core techniques such as neural network; and some suffer from determining a proper threshold while the user is unfamiliar with the domain knowledge. In this paper, we proposed the HIerarchical Shrinking decision Tree (HIS-Tree) algorithm to solve these problems. HIS-Tree uses the multivariae test derived from geometric mean measurement as splitting criteria to group minority examples together. By this way, HIS-Tree can avoid discovering rules dominated by the majority examples. Finally, as shown in the experiment, HIS-Tree can predict minority/interesting examples more accurately.
|Number of pages||9|
|Journal||WSEAS Transactions on Information Science and Applications|
|Publication status||Published - 2007 Jan 1|
All Science Journal Classification (ASJC) codes
- Information Systems
- Computer Science Applications