I am tabulating the results of a few studies in literature. Some studies have attempted classification and have indicated their results as accuracy in percentage, while some studies have done correlation analyses and have their results as correlation coefficients, represented as sub-unit values. I want to use a uniform percentage representation for all these studies. So is it appropriate to represent the correlation coefficients as percentage?
I am trying to read between the lines of your question when you write "classification . . . accuracy in percentage" and when you write "correlation analyses." I don't know what to make of a Spearman correlation in terms of sub-unit values. Tell us more about your "few studies in literature."
But the answer to your question is that classification and correlation are not compable (are qualitatively different), so a common mathematical treatment is not appropriate. As an example, if the correlation coefficient is negative 0.5 in one case and is positive 0.5 in another case, then any conversion of those coefficients to a percentage would not be meaningfully comparable to an "accuracy in percentage."
You may, however, want to read an explanation of r-squared.
Correlation is unit independent. (If there are 2 attributes A and B) if corr(A,b)>0 then it is called +ve correlation. if corr(A,b)
Classification accuracy is a percentage. Better to present a contingency tablein your results. Correlation coefficients are real numbers in the range -1 to +1.
- Badges
- Science topic
- Similar topics
- Data Mining
- Classification