Why I didn't have the same result when I calculate the balance accuracy in two different ways?
Method 1 :
balance_accuracy=recall_score(ytest,y_predicted, average='macro')
Method 2:
fpr1, tpr1, thresholds = metrics.roc_curve(ytest, y_predicted, pos_label=1.0)
spec=1-fpr1[1]
sens=tpr1[1]
bc=(sens+spec)/2
Thanks a lot
It was solved. In both of them , you should pass y_score not y_test. That's the problem.
If you pass y_score, then you will get the same results.
Be careful, in this case is just working for binary classification not multi-label.
Thanks a lot
Dear Jehad ,
For a dichotomized result we usually use the ROC curve (AUC). Recently we published a manuscript that compares pairs of samples, providing an estimation of accuracy without depending of cut-off as the ROC curve. We make available a Macro for download. We work with serological samples, however the methods can be applied in several areas. Maybe can help you!
DOI: 10.1039/C7AY00505A
https://www.researchgate.net/publication/314560508_Improved_assessment_of_accuracy_and_performance_indicators_in_paper-based_ELISA
https://www.researchgate.net/publication/315493970_m-Accuracy_Macro_in_Microsoft_ExcelR_to_estimate_test%27s_accuracy
Article Improved assessment of accuracy and performance indicators i...
Data (m-Accuracy): Macro in Microsoft Excel® to estimate test’s accuracy
- Badges
- Science topic
- Similar topics
- Linguistics
- Composition Studies
- Publication