T-test on two classifiers

hello mathworkers,

I have two classifiers classify the same dataset, run the experemnet ten times for each one on the same data, means I have for each classifer 10 accuracy result on same data , can i do the T-test to detrmine which classifier is best ?

Answer 1

[Star Strider]

I seriously doubt that a t-test or anything similar to it (even non-parametric tests) are appropriate to compare classifiers.

I refer you to the documentation section on classifier Model Building and Assessment for guidance. That in turn refers you to the testcholdout and testckfold functions that may be what you want.

I leave the rest to you.

[mohammad Alsajri]

thanks dear Star Strider

[Star Strider]

As always, my pleasure.

Answer 2

[Image Analyst]

Do you have ground truth for your data? If so, why not simply get the accuracy percentage for each classifier?