next up previous contents
Next: Efficiency of Learned Knowledge Up: Correct Prediction Previous: Correct Prediction

Prediction when the Training Set is Noisy

Instead of evaluating a system based on ``perfect training sets'', one should also measure a systems noise handling capabilities. Different levels of noise should be added to the training set, and the number of correct classifications with the new rules should be noted. If the noise added to the training set created inconsistencies, the resulting classification errors should be studied carefully as some systems for instance will create conflicting rules with no hints to which one to choose in a particular situation.



Helge Grenager Solheim
Sat May 4 03:30:02 MET DST 1996