Bayesian Logistic Regression Models for Credit Scoring by Gregg Webster

4.6.1 Cut-off probability of 0.3 
 
Logistic regression model 
Table 4.17
Classification table for the logistic regression model with cut-off probability of 
0.3.
Predicted 
Good 
Bad 
Actual 
Good 
1126 
196 
Bad 
100 
240 
Of the 1,662 applicants in the test set, the logistic regression model (Model 1) rejected 436 
and accepted 1,226 applicants. Of the rejected applicants, 196 (45.0%) are in fact good 
(Table 4.17). Therefore, 196 applicants are missed profits for the financial institution. Of 
the accepted applicants, 100 (8.2%) were bad - losses for the financial institution. Because 
the financial institution is only exposed to the applicants it accepted, the classification 
error is 8.2%. The overall classification error rate is 17.8%. The overall classification error 
rate gives a better indication since it includes applicants which represent missed out profits 
for the financial institution.
Bayesian logistic regression model with an informative prior
Table 4.18
Classification table for the Bayesian logistic regression model with informative 
prior and cut-off probability of 0.3.
Predicted 
Good 
Bad 
Actual 
Good 
1155 
167 
Bad 
106 
234 
Of the 1,662 applicants in the test set, the Bayesian logistic regression model with 
informative prior (Model 2) rejected 401 and accepted 1,261 applicants (Table 4.18). Of 
the rejected applicants, 167 (41.6%) are in fact good - this is missed out profits for the 

---

94 
financial institution. Of the accepted applicants, 106 (8.4%) were bad. This represents 
losses for the financial institution. The classification error rate realized by the financial 
institution is thus 8.4%. The overall classification error rate is 16.4%.
Bayesian logistic regression model with a non-informative prior