I receive the following comment from a reviewer : "I think that the authors could explain in more details the results. For instance, there is no discussion linking the specific properties of the datasets at hand and the performance of the classification methods".

To be honest I did not understand what the reviewer said. What are the specific properties of the dataset?

Is it possible just by looking at those specific properties to infer which classifiers will have a better performance?

Is it possible to explain the performance of the classifiers by the specific properties of the data sets?

Adding more details to my question. Suppose that classifier A has the following results in UCI datasets ( the performance measure is the accuracy in tem fold cross validation with 5 repetitions )

  1. Iris : 96%
  2. Wine: 93%
  3. Sonar:77%
  4. Pima: 75%
  5. balance: 84%
  6. Habeman:76%
  7. Breast: 96%
  8. Australian: 84%
  9. Ionosfere: 90%
  10. Lupus: 75%
  11. Bupa: 65%
  12. Transfusion: 77%
  13. Lawsuit: 96%

Classifier A presented good performance in Iris, Wine, Ionosfere and lawsuit. Is there any specific properties in commom with these datasets? Or Are they all just considered "easy" datasets?

  • $\begingroup$ I'm not familiar with lots of classifiers, so my answer is partial. Yes, you can infer what classifier can do by checking properties of the datasets. For example, if the data is intrinsically hierarchical, speech signal - images etc, I expect that a deeper structure would perform better than a shallow one . If the data is skewed, I check which classifier performs better on skewed datasets. If the data is a time series, I check if the classifier can treat consecutive features as dependent. $\endgroup$ – yasin.yazici May 26 '15 at 10:34
  • $\begingroup$ Thanks to @yasin.yazici, I decided to add more details to my question. I can understand the specific properties he mentioned. But what can I say about the specific properties of some well known UCI datasets such as íris, pima , bupa etc? $\endgroup$ – Jorge Amaral May 26 '15 at 13:39

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Browse other questions tagged or ask your own question.