Unbeaten classification dataset?

Looking for a dataset that meets the following criteria:

  • Multi-class , 3+ labels, preferably not binary.
  • Tabular data, not images.
  • Getting above 65-85% prediction accuracy is a challenge. In general, high prediction accuracy would be seen as a struggle.

I am hoping to prove some potentially novel classification methods. To date, I have been working with the classic Iris, Abalone, Sonar, and Kaggle Zoo datasets - while scouring Kaggle for more.

Anyone know of a toughie? Even if it is your own data.


Hi @layne.sadler,

I am not aware of a dataset like that. At least, I am not able to recall it as of now. However, here is a good list of datasets that you may want to look into:

I believe any dataset can be categorised as difficult, if we don’t have sufficient domain knowledge. So may be you should pick a dataset from a field about which you have no idea.