How to build a roc curve and do statistical analysis?

Question

I have 5 data sets with different sizes, for each dataset, I have selected 10 bootstrap samples of the same size as the database and for each sample, I have the values of true positives, true negatives, false positives, false negatives And the sample mean. Can I add up all sample values even though they have different sizes? What is the best way to construct the ROC curve of all databases using R? What is the best statistical test? I tried to use the pROC and ROCR packages, but they need the predicted values and classes (which are values I do not have, I have only the values of tp, tn, fp and fn).

Answer 1

I understand you are doing using some kind of predictive model on the bootstraps? The output of that model then somehow gets thresholded into Positive/Negative calls, when then are used for calculating TP, TN, FP, FN. If that's the case, the continuous output from your predictive model before thresholding is what should be used for pROC and ROCR.