Monotone Learning Datasets

The purpose of this dataset collection is to provide a test bed for monotone classifiers. The original data were collected from various resources, such as UCI and WEKA. A general preprocessing procedure was applied, taking the following aspects into consideration:

  1. Instances with missing values are ignored.
  2. Nominal values are transformed to numeric ones, according to the semantic.
  3. Features are normalized between 0 and 1.
  4. Datasets are formatted as .csv files with the class information in the last column.

We are consistently searching for real-world monotone learning datasets. If you have anything in mind, please do not hesitate to contact Weiwei Cheng.

Dataset # Instances # Features # Classes
breast cancer 278 7 2
car evaluation 1728 6 4
car mpg 392 7 36
cpu 209 6 4
den bosch 119 8 2
employee rejection acceptance (ERA) 1000 4 4
employee selection (ESL) 488 4 9
journal ranking 172 5 4
lectures evaluation (LEV) 1000 4 5
mammographic 830 6 2
social workers decisions (SWD) 1000 10 4

Download (in .zip-format)

These datasets have been used in:
  • Ali Fallah Tehrani, Weiwei Cheng, Eyke Hüllermeier
    Choquistic regression: generalizing logistic regression using the Choquet integral [pdf][bibtex][slides]
    Proceedings of the 7th Conference of the European Society for Fuzzy Logic and Technology (EUSFLAT-2011): 868-875, Atlantis Press
    Aix-les-Bains, France, July 2011 Nomination for Best Student Paper Award

  • Ali Fallah Tehrani, Weiwei Cheng, Eyke Hüllermeier
    Preference learning using the Choquet integral: the case of multipartite ranking [pdf][bibtex]
    IEEE Transactions on Fuzzy Systems, 2012.


Further information: