If you were to analyze how useful a set of attributes is to identifying a specific object out of a given set of objects, how would you do it? Most of the attributes are normally distributed, some could be rather random, while all objects are unique, but could still share the same values.

My naive approach is just to take the whole set of given data, look how many duplicates there are using those specific attributes, and calculate the probability for any given object finding a duplicate in any group. Maybe doing PCA afterwards to see which attribute contributes how much, but still.

I heard some answers like
>State Vector Machines
>Geometric Pattern Recognition
>This is a classification problem

And I have no clue if its them or me who is retarded. Is finding/recognizing unique elements really a classification problem?