The probabilities of logistic regression indicate how the certain the model is over predictions. if its 0.93 it means the model is 93% confident the label is 1 and 7% to be 0. or if the probability is 0.14 it means, 14% to be 1 and 86% to be zero.

In my case, I have 80% accuracy on the hold out data using 0.5 threshold over prediction. this value is calculated using AUC curve. However, if I set the threshold to be 0.9 to calculate the accuracy, the accuracy becomes 60%. Which is not so intuitive. I guess the probability over logistic regression are not necessarily the confidence or fitness of the model. I wonder if it is possible to get a confidence score over these values.

I have seen similar questions, where people have asked how to get confidence interval over prediction. This question is different, because the focus is on logistic regression.


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.