I used no data splitting in the development of models. A bit on data splitting and an example of PROC HPLOGISTIC in SAS that accomplishes it.
There are numerous measures that are examined when deciding on a threshold for classification.
It all begins with the definition of the confusion matrix.
Reproduce the final model for the develop data set
The three essential components for deriving measures of classification performance.
Some additional measures.
Some charts based on the measures.
A decision theory approach, introduction.
Bayes Rule.