Feature Shuffling

This method permutes(shuffles) a feature’s values and then measures the increases in error(or lowers accuracy). This method selects features which had increased accuracy as more important.
Features are selected as important if they highly contribute to prediction. I.e. A random permutation of feature's values will dramatically change Evaluation Metrics such as accuracy, MSE or ROC.


Measure of Predictivity is how much changing a future during Feature Shuffling effects model's quality, and is used in Feature Selection.