I'm looking for complex datasets for performing a comparison on machine learning in the cloud solutions such as Amazon Machine Learning, Microsoft Azure, IBM Watson, Google Prediction API, Rapid Miner and some others.

In particular I will try the blackbox prediction (and/or classification) capabilities offered by some of these systems, and compare it with state of the art results. SaaS such as Google Prediction API say they can pick the best algorithm for our problem; I'm analyzing how well do they perform in order to compare them with the closeness to state of the art results and some other further analysis.

It's hard to define complexity for this case, however I'm looking for datasets that have available results for at least 3 different algorithms, and where at least 1 has shown considerably bad performance (e.g. accuracy or error) in comparison with the best one. Size of the dataset will not be an issue. Following this idea complex would mean that there exists bad performing typical approaches that are considerably worst performing than the best solution.

Where should I look for that can have multiple algorithm results both well and bad performing? Any datasets recommendations?

PS: Naturally I would need datasets for supervised learning.

  • Have a look at kaggle.com - or MNIST, IRIS, Pascal VOC, ... Commented Jan 6, 2016 at 12:22
  • Thanks for the recommendation @moose I know Kaggle and the mentioned datasets, however the idea of the question is asking if someone know specific datasets that fit the characteristics I described. I would take a further look into Kaggle past competitions though.
    – Javierfdr
    Commented Jan 6, 2016 at 18:01

1 Answer 1


Have a look at this great tour-de-force study by Manuel Fernandez-Delgado at JMLR. They ran lots of classifiers against plenty of data sets from UCI. I believe you will find everything you need. hth


Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.