in a three class problem, how to find accuracy,sensitivity etc.. (For binary classification TP
Dear Aleka,
a popular approach is to divide the data into projections of different couples, please see article of Emidio Capriotti from 2008:
http://bmcbioinformatics.biomedcentral.com/articles/10.1186/1471-2105-9-S2-S6
Generally, if there isn't too many categories, I would recommend to construct N x N contingency table. A compression of performance into single metrics isn't easy, however, some applicable metrice can be cound in the following article written by Pierre Baldi in 2000 (see section "More than two classes"):
http://bioinformatics.oxfordjournals.org/content/16/5/412.abstract
Hope it helps,
Jaroslav
Dear Alaka Deepak,
You follow this paper:
https://www.researchgate.net/publication/303011477_Quality_Assessment_of_Web_Services_Using_Multivariate_Adaptive_Regression_Splines.
This paper focus on four class problem.
These are the ways to calculate the value of TP, TN, FP, FN
Here CON is confusion matrix:
TPvalue(1)=con(1,1);
TPvalue(2)=con(2,2);
TPvalue(3)=con(3,3);
TPvalue(4)=con(4,4)
TP=sum(TPvalue)
FNvalue(1)=con(1,2)+con(1,3)+con(1,4);
FNvalue(2)=con(2,1)+con(2,3)+con(2,4);
FNvalue(3)=con(3,1)+con(3,2)+con(3,4);
FNvalue(4)=con(4,1)+con(4,2)+con(4,3)
FN=sum(FNvalue)
FPvalue(1)=con(2,1)+con(3,1)+con(4,1);
FPvalue(2)=con(1,2)+con(3,2)+con(4,2);
FPvalue(3)=con(1,3)+con(2,3)+con(4,3);
FPvalue(4)=con(1,4)+con(2,4)+con(3,4)
FP=sum(FPvalue)
TNvalue(1)=n-(sum(con(:,1))+sum(con(1,:))-con(1,1));
TNvalue(2)=n-(sum(con(:,2))+sum(con(2,:))-con(2,2));
TNvalue(3)=n-(sum(con(:,3))+sum(con(3,:))-con(3,3));
TNvalue(4)=n-(sum(con(:,4))+sum(con(4,:))-con(4,4))
TN=sum(TNvalue/4)
Conference Paper Quality Assessment of Web Services Using Multivariate Adapti...
- Badges
- Science topic
- Similar topics
- Data Mining
- Classification