CODEWITHSUNDEEP
machine-learningnlp
user3799934
user3799934

Reputation: 105

How to know the which class the classified result is closer to in machine learning

Forgive me for being not clear in the title, I couldn't really phrase it in a nice way. But here's an example. Say I have a set of data with 3 classes: positive, neutral, and negative, and then I use them to train a classifier. Now I want to know that if that classifier predicts neutral, is it more of a positive neutral or a negative neutral.

So the most straightforward way I could think of is just checking the second highest possibility, or consider all possibilities but with different weight/coefficient.

Is there a term for this kind of problem? Any papers I should reach to? Thanks a lot.

Upvotes: 2

Views: 55

Answers (1)

Prune
Prune

Reputation: 77850

Any specific answer depends on the algorithm (Naive Bayes, 3-class SVM, spectral clustering, etc.) and implementation (analytics framework) you plan to use. The measure you're looking for is under the heading of "scoring", Most of these do give you programmatic access to the evaluation function.

Depending on your data space, finding the trained boundaries (so you know which is closer) can be more of a problem, since the boundaries may not be linear, or even have a good spatial representation.

As for papers to read ... (1) Without knowing more about your intended application, all I can suggest is to go read about Machine Learning and classification vs scoring; (2) Asking for resource references is one of the proscribe classes of question on StackOverflow.

Does this nudge you in a useful direction?

Upvotes: 2

Related Questions