How to set class_weight parameter for cost sensitive learning?

Question

I'm dealing with a binary classification problem with a balanced data set, however false positives are much more costly than false negatives. Let's just say that an FP is in general 3 times more costly than an FN and the response variable = 1 means a positive identification. How should I set the class_weight parameter in sklearn's RandomForest to reflect this ? From my understanding, I would say:

class_weight = {0:1.0,1:3.0}

I'm not sure if I understand this parameter correctly or should it be the inverse ?

Thanks.

Answer 1

\begin{array} {|r|r|} \hline - &(1) &(0) \\ \hline (1) &- &FP\\ \hline (0) &FN &-\\ \hline \end{array}

Rows elements are Y_true and column is Y_pred.

FP means, we predicted Positive and it came out False i.e. Class was Negative(0 here).

It means we don't want the Model to mis-classify the Negative class.
So, it implies, we will put a bigger penalty for the Negative class. Hence,

class_weight = {0:3.0.0,1:1.0}