Cross validation with a large number of folds (10+) is what I most commonly see recommended to reduce overfitting. To some extent, there will always be overfitting, even with cross-validation, because the hyperparameters are optimized to the training data which does not perfectly represent the testing data.