From fairlearn/fairlearn#1166, we have an inconsistency between liac-arff and pandas parser.

From the ARFF specs, the leading whitespaces are ignored if not between quotes. The pandas read_csv will include this space by default. E.g.

>>> import sklearn.datasets as skd
>>> d = skd.fetch_openml(data_id=1590, as_frame=True, parser='pandas')
>>> d.target
0         <=50K
1         <=50K
2          >50K
3          >50K
4         <=50K
          ...  
48837     <=50K
48838      >50K
48839     <=50K
48840     <=50K
48841      >50K
Name: class, Length: 48842, dtype: category
Categories (2, object): [' <=50K', ' >50K']

I am unsure that we can easily solve the issue because once read by read_csv, we don't have the information about the quotes anymore. I assume that the best that we can provide is to pass any additional keyword argument to read_csv to make it flexible enough.