 
  
  
  
  
 Next: Example
Up: Decision Tree Learning
 Previous: Why convert to rules?
 
-  dynamically creating  new discrete valued attributes   that
is true if that
is true if  -  sort examples according to the continuous attribute value
-  Identify adjacent examples that differ in their target
classifications
-  Generate candidate threshold midway between these points
-  Calculate the information gain of each candidate and pick best
-  dynamically created boolean attributes competes with others to
appear in tree
 
-  The value of c that maximizes information gain must be one of
these points
-  many other approaches
 
Patricia Riddle 
Fri May 15 13:00:36 NZST 1998