0
$\begingroup$

I'm having a classifier that tries to classify 5 different classes from a data set. It works pretty well in general, and when I'm plotting the confusion matrix almost all misses are 0, 1 or 2 max 3 except for the misses between 2 specific classes, which are >15. I'm a beginner with machine learning and I'd like to know if there's something I can do with this info. Of course I can tweak parameters, try other classifier etc, but I'm trying to remove chance or brute force from my method and would like to follow the logical steps.

So, what are some things to look after in a classifying situation in which the predictions are accurate between all classes, but 2 (the classifier mistakes one for the other often)

$\endgroup$
0
$\begingroup$

You should try to put more focus on misclassified classes.
There are 2 common ways to do that :
- increasing the proportion of classes misclassified during training
- use class weights in the loss to give more importance to the misclassified classes
I also advise you to do error analysis to check why those 2 classes have lots of misclassifications (is this because they are similar or because of a potential bug in data generation).

$\endgroup$

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.