In the class we talked about ROC curve, Mr. Lin said,
"The closer the curve follows the left-top corner, the more accurate the test".
I have some questions about this statement.
ROC curve plot "percentage of true positive" against
"percentage of false positive".
What if one method predict more negative than another method do?
Is it make sense to compare the ROC curves of different methods?
For example, method A predict 1 positive and 10000 negative, and
method B predict 10000 positive and 1 negative.It can easily achieve
the left-top corner for method A, but this method is not necessary a
good method.
To sum up, I do not know how to interprete the information from ROC curve.
Can we say R1 is better than R2 in the following picture?
http://www.csie.ntu.edu.tw/~r94152/ROC_question.PNG
Thanks.