I have a multiclass problem and for the class equal to 2 in the target I have some categorical columns with just one value.

For instance, is like for the observatuons with the target equal to 2, the column 1 is always equal to A, column two is always equal to 2 and so on... even though columns 1, 2 and others have multiple classes, for target equals to 2, the class is always the same. This lead to an accuracy of 100%.

Is this a sampling problem? What can it be and how would you solve it?


1 Answer 1


I think it would depend on the distribution of the classes with respect to those sets of features that are constant. As in, if you take all of the data points that have th? Are they very heavily proportioned for class 2, or are they randomly distributed amongst all classes as well?

Because from your description, it seems that you have found the class 2 will decide the values for that set of features, but the converse may not be true. So if those set of values will not decide class 2, then it means that the model will still learn something. whereas if that set of values is only for class 2, then there could be a causal reason for that. So perhaps those values for that attribute are supposed to decide class 2.

However, if that set of values for those attributes is attributed to other classes as well, then the model will still learn something. So I don't think it will pose that big of a problem at all.


Your Answer

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

Not the answer you're looking for? Browse other questions tagged or ask your own question.