2

Is there a collection of regression datasets for the purpose of benchmarking? For classification I found this paper (https://arxiv.org/abs/1708.03731) which has a collection of classification datasets (some of the UCI ML Repository). Any suggestions would be helpful.

The collection should include regression problems from different domains and should not be too easy or too hard (so that better models should get a relevant performance gain).

PhilippPro
  • 121
  • 2
  • Did you check the 'Related' questions that show up to the right? –  Dec 08 '17 at 16:36
  • Yes. I did not find any question that is similar to mine. There is one question asking about one (only one!) dataset, and the answer was also only one dataset (elemapi). So I do not get your point. I even think this is really a missing thing in the machine learning/data science/statistics community. – PhilippPro Dec 10 '17 at 08:41
  • What is your question, that is not answered by UCI ML Repo? – philshem Dec 10 '17 at 14:16
  • That these are arbitrary datasets. I want a collection of datasets to distinguish between different learners properly. For benchmarking. Not a dataset with 100 observation where linear model and random forest have the same performance. Something like in the paper I cited above. – PhilippPro Dec 11 '17 at 08:51
  • I've just added a bounty, and for me the UCI ML repo has mostly traditional ML regression datasets, uncorrelated, high-level feature data as input. I'm hoping to find a reference to a regression dataset that's best handled by neural network models (e.g. image, raw data, etc) – David Parks Oct 23 '18 at 14:54

0 Answers0