Multiclass data can be divided into binary classes. e.g. you have 3 classes of data named: A, B, C. You can do multiclass classification or you can divide them into the binary groups like: A-B, A-C, B-C then apply binary classification. I am doing SVM and interested what are the advantages/disadvantages of each of these classifications:multi vs binary.
My understating from your question is, it depends on data behavior and type of analysis. Another point is if your attaining equal accuracy then time complexity is a benefit than latter method.
Zeynab Mousavikhamene
https://www.quora.com/In-multi-class-classification-what-are-pros-and-cons-of-One-to-Rest-and-One-to-One
Have a look at : https://www.kaggle.com/c/otto-group-product-classification-challenge/discussion/13123Conference Paper Multiclass classifiers vs multiple binary classifiers using ...
Depends what the applications demand or require. For example if a medical application only requires to test positive or negative for a disease, then binary classification is sufficient, but if the application also requires disease sub-categories to be identified e.g. skin disease types, then multi-class classification would be necessary.
Difference is also related to speed of training/test process. If you have a lot of classes, using ovr (one versus rest) function shape is (nsamples, nclasses) and it is more computationally efficient instead of ovo (one versus one) where function shape is (nsamples, nclasses*(nclasses-1)) because you train less svm thus less comparisons.
- Badges
- Science topic
- Similar topics
- Data Mining
- Classification