10
$\begingroup$

I have a classification dataset where roughly 20% (maybe more) of the labels are incorrect. There is no way to know which labels are incorrect and no way to eliminate them in the future when further data is collected.

A method I see in dealing with this is to train an ensemble of classifiers and then take only the training data which matches the ensemble majority vote.

Are there any other algorithms/methods that are more resilient to data that is not labeled 100% correctly? Can we even treat this data as supervised learning? Is there anyway to trust the trained model or performance metrics such as accuracy and F1 score?

Thank you for the help.

$\endgroup$

2 Answers 2

6
$\begingroup$

This problem is known as "label noise" and there are a number of methods of dealing with it (essentially you need to include the possibility of incorrect labelling of patterns into the model and infer whether the pattern has been mislabelled, or actually belongs the wrong side of the decision boundary). There is a nice paper by Bootkrajang and Kaban on this topic, which would be a good place to start. This paper by Lawrence and Scholkopf is also well worth investigating. However, research on this problem has quite a long history, IIRC there is a discussion of this in McLachlan's book on "Discriminant Analysis and Statistical Pattern Recognition".

$\endgroup$
-2
$\begingroup$

If you have a considerable amount of data, I would suggest first to make tests with subsets of the data, like , so that you do not have all the mislabeled data for training. Maybe also using some technique with multiple weak classifiers helps.

$\endgroup$
1
  • $\begingroup$ Thanks for the answer. Can you please expand a bit on what you mean. How does breaking the data into subsets help and what to do with the subsets after you train them? Also, are you suggesting using an ensemble of classifiers? $\endgroup$
    – mike1886
    May 30, 2014 at 13:47

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.