2
$\begingroup$

Currently I am on a text classification project, the goal is to classify a set of CVs according to 13 classes. I use the bayes algorithm (ComplementNB), in my tests it is the model that gives the highest score. The problem is that I do not exceed 54% of recall, nor 50% of F1 score. I have a problem of class imbalance to begin with, but also my texts are similar and therefore my variables are not independent. looking at my confusion matrix, the model tends to classify the majority of resumes in majority classes. Would there be a solution to increase my F1 score and above all what should be done for the variable dependency problem? I want to clarify that the same text (CV) can have up to 4 different classes.

$\endgroup$
1
  • $\begingroup$ Welcome to DataScienceSE. Can you please edit the question to add a few details like number of instances, number of features, type of features (one-hot encoded words I assume?). And did you look at the performance on the training set and test set to check for overfitting? $\endgroup$
    – Erwan
    Jan 6 at 21:14

Your Answer

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

Browse other questions tagged or ask your own question.