Next: Summary
Up: Decision Tree Learning
Previous: Missing Attribute Values in
- Temperature, BiopsyResult, Pulse, BloodTestResults
- prefer decision trees that use low-cost attributes where
possible
- divide Gain by the cost of the attribute
- do not guarantee optimal cost-sensitive decision tree, bias the
search in favor of low cost attributes
- - robot domain
- , where is a constant that determines the relative importance of
cost versus information gain - medical domain
Patricia Riddle
Fri May 15 13:00:36 NZST 1998