As far as I understand logistic regression should be used for binary classification, so if you want to create a model that ask something like:

Will the market price rise or not?

House price is fair or not?

we are ok, otherwise you should use other kind of regression

Good question, I think it refers to the cost of algorithm error, in other words the error that will be incurred to use one model or another