Transactions on Machine Learning and Data Mining (ISSN: 1865-6781)
Volume 6 - Number 1 - July 2013 - Pages 19-42
Reduction of Distance Computations in Selection of Pivot Elements for Balanced GHT Structure
László Kovács1
1University of Miskolc, Department of Information Technology, Hungary
Abstract
For objects in general metric spaces, the generalized hyperplane indexing is one of the most widely used indexing techniques In the paper, some methods are presented to improve the quality of the partitioning in generalized hyperplane tree structure from the viewpoint of balancing factor. The proposed method to represent the elements in the target domain metric space is the usage of a distance matrix as it can model the distance relationship without any information loss. The efficiency of the partitioning depends on the appropriate selection of the pivot elements. As this method requires the knowledge of a great number of distance values between the objects, the paper proposes an interval-based distance value representation. Based on the test results, the given method dominates the usual techniques if the fullness factor of the distance matrix is between 3% and 35%.
Download Paper
(348 KB)