I'm implementing the perceptron algorithm and the voted perceptron algorithm for an assignment for university. For that I need to find some decent datasets.. I've tried the UCI repos and I've come up with:

that work quite nicely, the accuracy is pretty high for the linear separability. HOWEVER I need to find a third dataset to end my assignment, I've been searching everywhere but I can't seem to find anymore of them, the dataset has to have more than 1000 examples, and it has to be binary classifiable obviously..

Does anyone of you know where I could find such a dataset?

I've been trying the Adult dataset, but it's pretty much sparse.. far from being linearly separable..

NOTE: I can't use the kernel trick or any other higher dimensional paradigm.

  • $\begingroup$ There is a similar question on stats.stackexchange.com/questions/247057/… and in one of the answers they points taht there is a Iris Dataset with some of the classes been linearly separable $\endgroup$ Apr 7, 2019 at 19:52
  • 1
    $\begingroup$ Also this reddit question has a nice proposal for this kind of problem > "If you're just playing with data, another approach would be to find the separation that best separates a set of data, then remove the points or dimensions that violate the linear separation. It wouldn't be valid for publication, but it could still be useful for practicing your ANN work." $\endgroup$ Apr 7, 2019 at 19:56
  • $\begingroup$ This question is likely a better fit at Opendata.SE $\endgroup$
    – Stephen Rauch
    Apr 7, 2019 at 20:49
  • $\begingroup$ Could you use a synthetic data set? It should be very easy to generate your own dataset that is "almost linearly separable". $\endgroup$
    – G5W
    Apr 7, 2019 at 21:05


Your Answer

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

Browse other questions tagged or ask your own question.