When building a logistic regression, the probability of an observation belonging to a class can be calculated. But is there a way to determine how reliable that probability is? If the accuracy is of the model is low then how good is the probability?

  • 2
    $\begingroup$ I think this depends on how you would want to look at the "reliability" of the probability. Would you like a Confidence interval for the probability? Then maybe this will help: druedin.com/2016/01/16/predicted-probabilities-in-r or google some more on "confidence interval for predicted probabilitys" Or you kan just use the misclassification rate of your model to indicate the uncertainty. I hope this helps $\endgroup$ – B. Bram Mar 28 '18 at 14:09
  • $\begingroup$ Also possible duplicate of Measuring accuracy of a logistic regression-based model. $\endgroup$ – EdM Mar 28 '18 at 16:22

Browse other questions tagged or ask your own question.