Monash FCIT AI Consortium (MAIC)
Faculty of Computing and Information Technology, Monash University
* * * Seminar Announcement * * *
"DECISION TREES AND VARIABLE VALUED LOGIC RULES: A COMPARISON"
Dr. Xindong Wu
Software Development Dept, Monash
(xindong@insect.sd.monash.edu.au)
Wednesday 26 July 1995 at 2.00 pm
Room F6.47, 900 Dandenong Road, Caulfield East 3145
Abstract
========
Decision trees and variable valued logic rules have both been widely
used in machine learning, especially in attribute-based induction, in
which examples are described using a fixed collection of attributes.
This talk analyses the advantages and disadvantages of the decision
tree based learning algorithms (such as ID3, NewID and C4.5), and
compares these algorithms with the author's own HCV induction
algorithm, which generates rules in variable valued logic. In
addition to theoretical analysis, the talk provides a battery of
experimental results showing that the rules generated by HCV are more
compact and more accurate than the decision trees (or decision rules
derived from the decision trees) produced by ID3-like algorithms in
both artificial and real world domains. Peter Jackson (1990) claimed
that the reason many people use decision trees rather than rules is
that the ID3-like algorithms are simpler than other learning
algorithms. This talk will demonstrate that that argument is no
longer convincing.
----------------------------------------------------------------------
The MAIC seminar series is free and open to everyone.
Enquiries to: maic-request@fcit.monash.edu.au
URL: http://www.cs.monash.edu.au/~korb/maic-calendar.html