How to calculate accuracy,sensitivity and specifiity from confusion matrix and ROC graph?

How to calculate accuracy,sensitivity and specificity from confusion matrix and ROC graph?

Answer 1

[KALYAN ACHARJYA]

Just try to undestand the four parameters (in confusion matrix), you can compute rest parameters easily.

I would suggest you follow the following Tom paper (Best paper to understand ROC Curve)

http://people.inf.elte.hu/kiss/12dwhdm/roc.pdf

[KALYAN ACHARJYA]

If you know those four parameters (TP, FP..) then you can easily get confusion matrix and ROC curve.

For ROC, first we have to calculate specificity and sensitivity, then only you can draw ROC. (Dont think in reverse way, for graph we need maths data)

[KALYAN ACHARJYA]

ROC Curve:

In biomedical image analysis, another very easy term to evaluate the performance of disease classifier or image identifier (object recognizer) is receiver operating characteristics (ROC) curve [R], all domain called ROC space. To calculate those parameters, two approaches follows lesions based and pixel based, though to find the pixel based result in highly texture and colors variant images is not so easy, on such cases lesion based approach can be used. Though ROC curve are drawn from quality parameters, The ultimate performance evaluation of disease classifier is done by ROC curve (1-specificity vs. Sensitivity) by changing the threshold values in coordinates space (0, 0) and (1, 1). The ideal cases for ROC curve is points/curve should be most nearest/towards to north-south corner (0, 1) as far as possible. The worst case be the near to red dotted line, which is random 50% probability to truly recognize the presence of disease. Already mentioned that, the ideal case is curve connecting these three (0, 0), (0, 1) and (1, 1) points, in figure shown the ROC curve of the retinal vessel extraction results for understanding purpose.

The quality parameter is Area under Curve (AUC): the maximum area covers by curve from east-south corner; the more area in results represents better results as compare to others. Its values varies from 0.5 to 1, if ROC curve covered the complete area of ROC space, then AUU is 1, which is ideal case. The worst case is 50% area covered, curve crosses through red dotted reference line as shown in figure.

[Aakash Mandhan]

Hello Sir, please send me the pdf file of matlab code for this post. [e-mail]

[Anjoo Kumari]

Sir can you give me matlab code plz sir