Just wondering if anyone can help clarify this.
For AUC analyses does the number of items on the scale impact how high the classification accuracy is? For example, will a 10 item measure always outperform a 2 item measure simply because of the fact that it has more items? If you can direct me towards any research that discusses this it would be helpful.
I think that research will be helpful in your problem
Article 100% Classification Accuracy Considered Harmful: The Normali...
Good luck.
Dear Martina Faitakis ,
I would say that compute AUC with only few items is not fair.
I would suggest to use at least 15-20 items for ROC curve computation, in order to obtain a finer curve. In this case, AUC has a useful meaning.
Dear Martina Faitakis
This stack exchange discussion might be useful, have a look
https://stats.stackexchange.com/questions/284389/auc-vs-accuracy-for-model-accuracy-evaluation
Kind Regards
Qamar Ul Islam
- Badges
- Science topic
- Similar topics
- Data Mining
- Classification