Metthew Correlation Coeffecient

Hi All

I would like to calculate Metthew correlation coeffecient for four class of data for example

N1=25,36,98,78,98,53....n, 
N2=45,12,25,36,17,45....n, 
N3=12,13,24,25,51,62....n and 
N4=13,15,61,17,81,28....n

any one could you kind enough to provide information or code to calculate Metthew correlation coeffecient by using MATLAB code. thank you very much in advance for your help and assistance.

Answer 1

[the cyclist]

I didn't find any explicit calculations of Matthews correlation coefficient (MCC) in either MATLAB or the File Exchange. However, here are a couple things that might help you. First, MATLAB will calculate the confusion matrix, with the confusionmat() command. Using that, and the formula for MCC that can be found here:

http://en.wikipedia.org/wiki/Matthews_correlation_coefficient

you might be able to calculate it yourself.

Also, searching for "confusion matrix" at the FEX turned up the following contribution:

http://www.mathworks.com/matlabcentral/fileexchange/21212-confusion-matrix-matching-matrix

There is mention in there of the calculation of "true positives", etc., that might be helpful to you.

Good luck!

[Malde Gorania]

Thank you Sir

Answer 2

[John]

Hello, I'm just wondering if you have soloed this problem yet. I'm looking for answers about similar question, but one of my problem is, as described by Wiki, MCC is used in binary classification. So in the case of multiple classes, I wonder how MCC should be calculated.

Thanks.

Sincerely,

[Eric T]

MCC is technically only defined for binary problems. Multiclass evaluation is an active area of research (e.g., http://arxiv.org/abs/1008.2908 is a good source of references, if not the best paper).

Answer 3

[TaeKeun Yoo]

The equation from https://en.wikipedia.org/wiki/Matthews_correlation_coefficient

function [mcc] = matthews_confusion(confusion)

%confusion : confusion matrix input

n = length(confusion);

upper=0;

for k=1:n

for l=1:n

for m=1:n

upper = upper + confusion(k,k)*confusion(l,m) - confusion(k,l)*confusion(m,k);

end

end

end

down1=0;

down2=0;

for k=1:n

down1_1=0;

down1_2=0;

for l=1:n

down1_1 = down1_1 + confusion(k,l);

end

for k_dot=1:n

if k_dot ~= k

for l_dot=1:n

down1_2 = down1_2 + confusion(k_dot,l_dot);

end

end

end

down1 = down1 + down1_1*down1_2;

end

for k=1:n

down2_1=0;

down2_2=0;

for l=1:n

down2_1 = down2_1 + confusion(l,k);

end

for k_dot=1:n

if k_dot ~= k

for l_dot=1:n

down2_2 = down2_2 + confusion(l_dot,k_dot);

end

end

end

down2 = down2 + down2_1*down2_2;

end

mcc = upper/(sqrt(down1)*sqrt(down2));