|
Neural
Networks |
Decision
tree rule induction is a method which aims to construct a set of rules
which will classify objects from knowledge of a training set of
examples whose classes are previously know. The process of
classification can be simply defined as the task of discovering rules
or patterns from a set of data. The objectives of any classification
task is to at least equal and essentially exceed a human decision
maker in a consistent and practical manner. The success of any
classification algorithm can be measured in terms of accuracy which is
determined by it's performance on an unseen set of data (test set) and
it's speed, the time taken to learn a set of rules or discover
patterns from a training set of data. Providing the training set is
randomly sampled from the given data set, the accuracy of the test set
can be seen to be an un-biased estimate of performance.
The method is based on recursive partitioning of the sample space and
defines classes structurally by using decision trees. The task of
constructing a tree from a training set is known as tree induction. A
number of algorithms exist such as ID3, C4.5, C5, CHAID and CART which
use general to specific learning in order to build simple knowledge
based systems by inducing decision trees from a set of examples.
Selected Reading
[1] J.R. Quinlan, Induction Of Decision Trees. Machine Learning 1,
Kluwer Academic Press, 81-106 (1986)
[2] J.R. Quinlan, Probabilistic Decision Trees. Machine Learning
Volume 3 : An AI Approach. Eds. Kockatoft, Y. Michalshi, R. pp
140-152, 1990
[3] J.R. Quinlan, Improved Use of Continuos Attributes in C4.5,
Journal of Artificial Intelligence Research, 4 pages 77-90, 1996.
Web Sites
Ross Quinlan - AI Department, CSE |
