The evoked performance measures in this section, served to evaluate the validity and the discriminant power of the studied models. From the previous results, we remark that the most
banks’ practiced model M1 seem the least successful model
once applied to non customers borrowers’ data, this confirms
the difference between the two studied subpopulations. We
also remark that models M5 and M6 might be a good
classifier. However, models M3 and M4 seems to be the more
suitable models for the prediction of non customers behavior to
pay back loan. These last one are the best predictive models
because their constant is calculated from the non customers
learning sample independently of customers sample, what
supposes their importance in the reliability prediction of the
target variable kredit and confirm the existence of a certain
link between the two subpopulation Ω and Ω∗. Model M5
possesses the most raised rate of Type II error, this model is
considered as careful but its use can lead to a loss of reliable
borrowers.