ROC result interpretation

Question

I am using WKEA for classification. I am using two algorithms adaboost and RBFNetwork. Surprisingly both of these algorithms are not performing well on my data and giving following results:

                  Adaboost       RBFNetwrok
      Precision :  0               0

      Recall     : 0               0

      F1-score   : 0               0

     Accuracy   : 91.36           91.36

     ROC_AUC   : 77.11         64.26

We can see that both of the algorithms are giving same value for 4 metric (precision, recall, f1-score, accuracy), but they are giving different result for ROC_AUC.

I am not able to understand,how it is possible? Am I doing error?

Please let me know.

Answer 1

This is absolutely normal. The AUC is integrated over all thresholds, while accuracy has been measured on a single threshold. This means the ROC curves can look quite different, with different AUC, but still share a common accuracy at some threshold (red circle):