The method of using the F-test for comparing two classifiers is somewhat loosely based on Looney's work.
Thus method can be used to compare two or more classifiers. And, in the context of the F-test, our null hypothesis is that there that there is no difference between the classification accuracies
A null hypothesis is a type of hypothesis used in statistics that proposes that there is no difference between certain characteristics of a population. In this case, we are trying to prove that there is no difference between the classification accuracies of the multiple classifiers, therefore, proving the null hypothesis.
This work is a small test to demonstrate the F-test and described in the MLxtend paper. It uses the Iris dataset. To conduct the test, we compare the accuracy of 5 different models, and then test the null hypothesis:
- SVM Linear
- SVM RBF
- Linear Discriminant
- KNN
- Perceptron