FriedmanTest<T>

Models & Types · AiDotNet.Evaluation.Statistics

Friedman test: non-parametric test for comparing multiple classifiers across multiple datasets.

Implements: IMultipleSampleTest<T>

For Beginners

The Friedman test is the standard statistical test for comparing multiple machine learning algorithms across multiple datasets. It’s recommended by Demsar (2006) for ML algorithm comparison.

  • Ranks each algorithm within each dataset
  • Tests if the average ranks are significantly different
  • Non-parametric: doesn’t assume normal distribution of scores

How It Works

Typical workflow:

  • Run k algorithms on n datasets (e.g., via cross-validation)
  • Use Friedman test to see if there’s any significant difference
  • If significant, use post-hoc tests (e.g., Nemenyi) to find which pairs differ

Constructors

Constructor Summary
FriedmanTest()

Properties

Property Summary
string Description { get; }
string Name { get; }

Methods

Method Summary
StatisticalTestResult<T> Test(T[][] samples, double alpha) Performs the Friedman test.