Thanks for the explanation!! However, since the cut-off value changes during each step in the cross-validation, in the last example of LOOCV, what could be the best cut-off value? How can we calculate that??
at 18:28 After understanding that specificity is 80 per cent, how do we choose the best cut off point? the same question is for k-fold. when following the steps(at 11:37), does the cut-off change in each step? If yes, suppose that when we are done with the k-fold (at 12:00) and we receive new data what cutoff will be used to evaluate that new data point?
This is the best explanation of ROC curves I have ever seen in my life. Thank you very much.
Thanks for the explanation!! However, since the cut-off value changes during each step in the cross-validation, in the last example of LOOCV, what could be the best cut-off value? How can we calculate that??
Use all (training) data to determine the best cut-off value to be used. LOOCV is only used for cross-validation, not to determine the best cut-off.
at 18:28 After understanding that specificity is 80 per cent, how do we choose the best cut off point? the same question is for k-fold.
when following the steps(at 11:37), does the cut-off change in each step? If yes, suppose that when we are done with the k-fold (at 12:00) and we receive new data what cutoff will be used to evaluate that new data point?
Yes, the cutoff value changes during each step in the cross-validation. I would select the final cutoff based on all training data.