Some ML algorithms use n$-$fold cross$-$validation $($CV$)$ over the training set with the purpose to:

Resample the training dataset for randomization.

Optimize over the training dataset.

Establish the correct accuracy value.

Remove the use of the random seed.