use of nearest neighbor and hierarchical methods (e.g., decision
tree and agglomerative hierachical clustering) in both
classification and clustering would be able to support this
viewpoint. In addition, both classification and clustering
have to attack similar issues (e.g., feature selection, scalability,
and missing value), and many solutions to the issues
can be used in both tasks without too much modification.
For example, when computing the goodness of an attribute
for classification or clustering, the difference is that the former
usually only considers class information while the latter
will take all attribute information into account. As a demonstration,
Fig. 3 of section 4 shows that a classification tree
(or decision tree) using the class information could be the
same as a clustering tree (or monothetic divisive tree, in numerical
taxonomy literature) without class information.
Shaded similarity matrix is one of the oldest graphic
techniques that has long been used in hierarchical cluster
analysis [23, 14, 30, 7]. Based on the above unified viewpoint,
we believe it could also be used in the task of classification
visualization. We will focus on the use of shaded
similarity matrix for visualizing two popular classification
methods: nearest neighbor and decision tree. We will also
explore how to attack the scalability problem using ensemble
classification and sampling techniques.
use of nearest neighbor and hierarchical methods (e.g., decisiontree and agglomerative hierachical clustering) in bothclassification and clustering would be able to support thisviewpoint. In addition, both classification and clusteringhave to attack similar issues (e.g., feature selection, scalability,and missing value), and many solutions to the issuescan be used in both tasks without too much modification.For example, when computing the goodness of an attributefor classification or clustering, the difference is that the formerusually only considers class information while the latterwill take all attribute information into account. As a demonstration,Fig. 3 of section 4 shows that a classification tree(or decision tree) using the class information could be thesame as a clustering tree (or monothetic divisive tree, in numericaltaxonomy literature) without class information.Shaded similarity matrix is one of the oldest graphictechniques that has long been used in hierarchical clusteranalysis [23, 14, 30, 7]. Based on the above unified viewpoint,we believe it could also be used in the task of classificationvisualization. We will focus on the use of shadedsimilarity matrix for visualizing two popular classificationmethods: nearest neighbor and decision tree. We will alsoexplore how to attack the scalability problem using ensembleclassification and sampling techniques.
การแปล กรุณารอสักครู่..
