Re: DM: How to improve the decision tree's generalization performance

[John Day]

Hi,

I suggest you read Quinlan's own book on C4.5 (J.R. Quinlan (1993). 
C4.5: Programs for induction. Morgan Kaufmann). "Pruning" is the
method most often used to optimize the generalization error in 
decision trees. "Cross validation" is another technique used to
improve estimates of the generalization error.
Both techniques are discussed in the book.

Hope that helps.

John Day/CSI
At 12:32 PM 9/27/99 -0600, fardin akbaryan wrote:
>    In my research, I am using the decision tree( C4.5) as a 
>classifier.
>
>The available training data set for my research do not cover all the
>input space which is common in classification methods. I have 
>observed
>that the performance of the tree on the unseen patterns is poor. In
>another word, my classifier lacks the reliability in terms of the
>generalization. As you know there are many methods and theories to
>improve the classifier's capability for the generalization. But I 
>need
>to have a method specifically for a decision tree.  I'll really
>appreciate
>if you give me a clue about this matter. Thanks in advance for your
>help.
>
>Fardin
>--
>
>
>Fardin Akbaryan
>Chemical and Petroleum Eng. Department
>University of Calgary
>2500 University Drive NW
>Calgary T2N 1N4, Alberta, Canada
>Tel: (403) 220-7409
>Fax: (403) 284-4852
>