What is a good metric for measuring the level of multicollinearity?

For classification models, techniques like stratified sampling are often preferred because they maintain the class distribution in the training and testing datasets