Hi Tapsi,

Fault detection and diagnosis (FDD) rate does not depend to the method of classification. For all classifier you can use the same method for assessing the performance.

In the literature, after designing the classifier and parameter tuning, you should apply it on a test dataset. Then the true label (class/fault) can be compared with the prediction of classifier. 

there are different approaches for performance evaluation of a classifier as a fault diagnosis system. The comprehensive one is confusion matrix. Based on confusion matrix different criteria can be calculated. 

Please find the attached paper. It is very interesting. 

If you have any problem with the concepts and implementation, do not hesitate to contact me. 

Thank you so much Askarien ma'm !

I have to implement the things on Matlab.

Please help me to understand the concept of Confusion matrix?

Also, please tell me how to implement the idea to calculate fault recognition rate in Matlab.

Pls give me your email id, if possible.

assume that the label is M*N (M observation and N class)

labelY: true Label

labelY: predicted label by classifier

N:number of classes

%change lable  to vector

llY=zeros(1,M);

llYP=zeros(1,M);

for i=1:N

[r,c]=find(lableY(i,:)==1);

llY(c)=i;

r=[];C=[];

[r,c]=find(lableYP(i,:)==1);

llYP(c)=i;

r=[];C=[];

end

confMat=zeros(N);% Global confusion matrix

% raw:predicted fault , column:true fault

for i=1:M

confMat(llYP(i),llY(i))= confMat(llYP(i),llY(i))+1;

end

tp=diag(confMat);

fp=sum(confMat,1)'-tp;

fn=sum(confMat,2)-tp;

F1=2*tp./(2*tp+fp+fn);

my Email: [email protected]