Exploratory and Inferential Analysis of Benchmark Experiments

Benchmark experiments produce data in a very specific format. The observations are drawn from the performance distributions of the candidate algorithms on resampled data sets. In this paper we introduce a comprehensive toolbox of exploratory and inferential analysis methods for benchmark experiments based on one or more data sets. We present new visualization techniques, show how formal non-parametric and parametric test procedures can be used to evaluate the results, and, finally, how to sum up to a statistically correct overall order of the candidate algorithms