Efficient Construction of Regression Trees with Range and Region Splitting

• ISHII Hiromu

  Department of Information Science, University of Tokyo

• MORIMOTO Yasuhiko

  Tokyo Research Laboratory, IBM Japan Ltd.

• MORISHITA Shinichi

  Institute of Medical Science, University of Tokyo

Bibliographic Information

Other Title

• 区間・領域分割を用いた Regression Tree の構成

Abstract

In recent years data mining has attracted many researchers among both artificial intelligence and database communities. Construction of regression trees is a topic of data mining. A regression tree is a rooted binary tree such that each internal node contains a test for splitting tuples into two disjoint classes. The mean of the objective attribute values at the leaf is used as the predicted value of the tuple. To test a numerical attribute, traditional methods use a guillotine-cut splitting that classifies data into those below a given value and others. In this paper, as an alternative of guillotine-cut splitting, we consider a family R of grid-regions in the plane associated with two given numeric attributes. And we propose to use a test that splits data into those that lie inside a region R and those that lie outside. Some experimental results showed that regression trees constructed through our method have higher accuracy than those through guillotine-cut splitting.

Journal

• IEICE technical report. Artificial intelligence and knowledge-based processing

  IEICE technical report. Artificial intelligence and knowledge-based processing 97 (415), 57-62, 1997-12-02

  The Institute of Electronics, Information and Communication Engineers

Keywords

• data mining
• numerical value prediction
• regression tree
• range and region splitting
• convex function

Details

• CRID: 1573950402150458112
• NII Article ID: 110003186761
• NII Book ID: AN10013061
• Text Lang: ja
• Data Source:
  • CiNii Articles