Next: Missing Attribute Values in
Up: Decision Tree Learning
Previous: Other Measures for Picking
- if
, then 
is undefined or
very large - to avoid selecting attributes on this basis
- calculate
of each attribute - calculate only on attributes with above average
- choose best
- many other evaluation functions
- distance metric Lopez de Mantaras, 1991
- distance between our partition and the perfect partition
- not biased by number of values for an attribute
Patricia Riddle
Fri May 15 13:00:36 NZST 1998