Wangdi Kinzang. Classification of loan borrowers of national pension and provident fund of Bhutan : a case study. Master's Degree(Information Technology). King Mongkut's University of Technology North Bangkok. Central Library. : King Mongkut's University of Technology North Bangkok, 2009.
Classification of loan borrowers of national pension and provident fund of Bhutan : a case study
Abstract:
The National Pension and Provident Fund (NPPF) of Bhutan has introduced member
financing schemes such as Housing and Education loan to its members since 2003.
The number of members applying for the loans is growing every year. Granting a
loan is becoming an important task for Loan Officials to avoid potential risks in the
future. However, the occurrences of repayment defaulters and Non Performing Loan
(NPL) as well as a number of cases that the mortgages have to be ceased have been
increasing significantly. With an aim to investigate a suitable classification model as
a decision tool for loan officials to judge future loan applications, the thesis
comprehensively studies five different algorithms, namely MLP, RBF, SMO, J48 and
RF. Applying five algorithms directly to the original dataset provides unacceptable
accuracy of bad borrowers detection at 1.1%. This is because NPPF dataset is
imbalanced and the five algorithms solely cannot learn very well from marginal
proportions of bad borrowers. Two data manipulation techniques, i.e., SMOTE and
segmentation, are additionally investigated. The experimental results indicate that the
performance is highly increased when SMOTE and segmentation techniques are both
applied. The best model can accurately classify applicants at 96.76%, with 81.93%
true positive detection of bad borrowers.