Some ML algorithms use n-fold cross-validation (CV) over the training set with the purpose to:
a) Resample the training dataset for randomization.
b) Optimize over the training dataset.
c) Improve the test accuracy.
d) Decrease the computational complexity.