# Accept any suggestion to create training data from correlation matrix to find odd one out to identify difference in variation

I have N time varying feature vectors obtained by recording different parameters over time.This results in N*N similarity matrix which contains one to one correlations value for each feature. We need to consider only the upper triangle matrix since it's symmetric, so the correlation value for one feature is low corresponding to all other feature. Is there any learning method that can identify that feature autonomously. For Example the matrix looks something like this

             Feature 1   Feature2 Feature 3 Feature 4
Feature1  1           0.91      0.81       0.44
Feature2  0.91          1       0.98       0.31
Feature3  0.81       0.98       1          0.32
Feature4  0.44       0.31       0.32        1


So we know Feature 4 is behaving differently . Any learning method that can learn this difference and identify the corresponding feature.Sorry, if this is a very trivial question, I'm new to data analysis.

• You've already noted that feature 4 is less correlated to the other features. You've already "learned" that difference by computing correlations. What else do you mean? Mar 5 '16 at 9:31
• The original data that I'm handling is not that small ,so instead of manually noticing the difference can we have a function that returns feature 4 in this case by understanding the variation. Thanks for answering , i would appreciate any help to start with this work,. Finally I want to have a function that returns the index of least correlated data like this index_least_correlation = func(Matrix) Mar 5 '16 at 9:38