Please take a moment to complete this survey below
Library's collection Library's IT development CancelBank X is a leading financial service company and already active in over
120 countries, including Indonesia. One of main services of Bank X is SME
Lending. Because of the tight competition in the same services, Bank X is
challenged to provide faster loan application process but has lower risk. A better
credit scoring model is needed because it can provide faster application process and
lower risk of default payment.
A poor credit scoring model has a poor power to detect the defected
lenders. Bank X has higher risk because of the poor credit scoring model. Through
this thesis, the current credit scoring model of Bank X will be validated. With KSScore
of 0.154, the current credit scoring model is not good anymore because it
cannot separate the non-default and the default applicants.
Both classical logistic regression and AUC (Area Under Curve)
optimization using Nelder-Mead Algorithm perform well in improving the model,
but logistic regression still better in some aspects. AUC optimization model has
higher AUC than logistic regression model but lower errors. Although AUC
Optimization better than logistic regression, the model still has a large error. The
model rejects about 81.94% of all applicants with score below the threshold.
By doing a deep analysis of applicants’ population, it is possible to reduce
the potential error of the model. By knowing the true default applicants’ population,
the population can be separated into two groups. The type I error of group one is
69.95% and 92.97% for group two. Bank X will get a lower potential of lost if they
approve the loan from group two applicants rather than from group one.