Next: Statistical Questions in Machine
Up: Evaluating Hypothesis and Experimental
Previous: Four Important Sources of
- Good statistical test should not be fooled by these sources of
variation.
- To account for test-data variation and the possibility of random
classification error, the statistical procedure must consider the size
of the test set and the consequences of changes in the test set.
- To account for training-data variation and internal randomness,
the statistical procedure must execute the learning algorithm multiple
times and measure the variation in accuracy of the resulting classifiers.
Patricia Riddle
Fri May 15 13:00:36 NZST 1998