Hello everyone,
I have a logistic regression problem at my disposal. I need to build a model such a way that it predicts/classify accurately. The data I am having is having the response variable values of unsuccessful event is too much. That is the ratio of 0s (84 % (26880 values) is much higher than ratio of 1s (16 % (5120 values)).
When I applied usual logistic regression to this data , I obtained around 82% of accuracy sensitivity but specificity was exact 0. Also when I applied Logistic regression using Firth's method the specificity did not improve much, it came around 44 percent. Which is not desirable.
I need to know what else can be done in this kind of data you get and you need to apply logistic regression ( or need to predict a category ) ?
Is there any well proven method already to handle this kind of data other than Firth's method?
Regards,
Irshad