I am working for a multiclass classification problem say 0,1,2,3,4,5. where say class 2,3,4 are being misclassified a lot. So, I want to train the model twice. At first, for all the six classes and then again for those 3 classes only. Then I want to merge these two results and have a final classification result for all the six classes. But I do not know how to. Do you have any suggestions? I am using Keras to train and test the model.
Asked
Active
Viewed 114 times
1 Answers
0
You can try to train two or even three classifiers. First that splits the data into (0, 1, 5) and (2, 3, 4) classes, and next one classifies data given that they are in first or second "bin".
This however might not solve the problem if underlying issue is the data. You must ask yourself if you have enough positive and negative examples for said classes. If no you should get more data or try to augment it. Secondly you should try to think if you can reasonably expect the model to give correct answers.
Mateusz Puto
- 78
- 7