I working on a research paper in Machine Learning, the paper is regarding a classification problem of which I have achieved 98% accuracy. However some other paper working on the same dataset claims to have achieved 100% accuracy. Is it still possible to use my results as state of the art results and publish my paper or should I just search for another dataset? If yes then do I mention the previous results paper which got 100% accuracy as previous work, or should I just disregard it?

  • 5
    Of course, you have to name all previous work you are aware of, especially if so relevant to your question. I am also a bit sceptical about 100% claims. You should verify how they did the comparison and if they actually had a set of unseen data with which they compared. – Captain Emacs Mar 22 '19 at 15:57
  • Any suggestions on how to verify previous work claims given that algorithm details/implementation were not provided – s.ali Mar 22 '19 at 16:26
  • @s.ali - that might be a good topic for an additional question. In short: do the best you can, then in the paper, you can say "there is a claim of X accuracy, we attempted to verify X, but the authors of X didn't provide any details, so the best we could get by making reasonable assumptions was Y < X." – cag51 Mar 22 '19 at 16:28
  • What do you mean by "search for another dataset"? – henning -- reinstate Monica Mar 22 '19 at 16:54
  • To find a new dataset and get state of the art results on that dataset, since the current one already state of the art results were achieved. – s.ali Mar 22 '19 at 16:59

Since you are aware of the other work you should report it. But you should probably also try to verify its claims.

Whether your own work is worth publishing or not depends on some things. To be valuable, it needs to advance the state of the art on some dimension and accuracy is only one. If you can achieve "really good" results in half the effort/time/cost, you probably have something worthwhile. It might make the difference between something that is feasible or something that is not.

But it would be a serious breach to ignore the other work.

| improve this answer | |
  • Any suggestions on how to verify previous work claims given that algorithm details/implementation were not provided. – s.ali Mar 22 '19 at 16:25
  • 1
    +1 for "it would be a serious breach to ignore the other work" - simply not an option. – Captain Emacs Mar 22 '19 at 16:27
  • 1
    Pretty hard. Without details the claims are just claims, not established fact. You could even cite it that way. "Claims are made in X that ..." – Buffy Mar 22 '19 at 16:27
  • 4
    One way to make your work valuable would be to make the details and implementation publicly available. – Patricia Shanahan Mar 22 '19 at 19:39
  • @Buffy I would push even harder. "X claims to achieve 100% accuracy, but these claims cannot be verified, because they do not provide a detailed description or implementation their algorithm." – JeffE Mar 24 '19 at 10:35

Your Answer

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

Not the answer you're looking for? Browse other questions tagged or ask your own question.