|
In my dataset, I have 2 labels, positive and negative. Most samples belong to only one class, either positive or negative. A small fraction of samples take both labels i.e. both positive and negative. Thus, making it a multi-label classification problem. Usually, this could be solved in many different ways including problem transformation or using multi-label classifiers. My question is, can I consider the samples that take both labels as a third class and consider this as a multi-class classification problem with 3 classes? As far as I understand, this method is not preferred in multi-label classification scenarios where the no. of individual labels is large. That is, if n is the no. individual labels, then a sample could take any of the 2^n - 1 combination of labels. Hence for large n, opting for a multi-class classification (with 2^n - 1 classes) could be a bad idea. But, in my case n=2, which is a trivial case. Hence, I am considering multi-class classification as an option. Please advise on this and let me know if I am right. Also, if possible please point to literature where such trivial case of multi-label classification problems are dealt with. |
|
Performance wise I don't see the issue but it will help if you can a little more about your third class. My guess is your third class must have some distinction else you can reduce this to a 2 class problem. i.e just solve for either positive or negative. |
