Finding odd one out

    down vote


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
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.