I built a Random Forest model for a binary classification problem.Both the classes in the target variable are balanced. My main class of interest is 'class 1'.

False negatives are more costly to me, so it would make more sense to reduce false negatives by optimizing the recall.

I read online that I can assign weights to my classes to penalize false negatives ( basically misclassification of class1) or optimize the threshold using a precision-recall curve to improve the recall for class 1.

Im my case ( where both classes are almost balances) which method would be better?Can I still assign class weights to penalize false negatives?If yes, how do I determine what weights to apply to each class?or optimizing the threshold is better?

Here is my classification report:

enter image description here

  • $\begingroup$ This question is a little under-specified. What do you mean by 'optimize'? If you want to maximize recall, simply build a simple classifier that labels all predictions as 1. That probably isn't what you want - which means your post needs more detail about what outcome you'd like. $\endgroup$ – tom Jun 6 '19 at 2:42
  • $\begingroup$ If you can create a mathematical definition of how costly a false negative then that can be the weighting for your classifier $\endgroup$ – Anonymous Emu Jun 6 '19 at 8:17

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.