Johan Oudinet. Choosing between two learning algorithms. CSI Seminar May 2007

In the research of machine learning algorithms for classification tasks, researchers have few training data for evaluating their algorithms. They have to apply test heuristics, but an heuristic contains errors.

We analyze two error types for five statistical tests: the probability of detecting a difference when no difference exists (type I error), and the probability of not detecting difference between algorithms when it exists (type II error).

• oudinet-statistical-test-slides.pdf

to top