|
Abstract:
|
This tutorial provides an introduction to statistics for Evolutionary Computation. We will break down some experimental analysis and statistical myths in EC. In the introduction, we cover how to compare two experimental designs against one another fairly and how to compute confidence intervals suitable for graphing and testing. We cover the Student's t test and its binomial approximation. In the more advanced part of the tutorial, we cover non-parametric pairwise tests, the costs of using the ranked t test, confidence intervals about the median, and measuring effect size using Cohen's d'. We end with a discussion of the statistical perils of doing science by computer, and common issues with stats in EC specifically. We also highlight the Bonferroni correction and introduce ANOVA to do multiple- factor and multiple-dimension testing.
|
