Development of Loan Default Prediction Model for Finance Companies in Sri Lanka - A Case Study

Finance Companies (FC's), play a pivotal role in the economy of Sri Lanka, by serving the under banked and non-banked segments of the society. The business model entails lending to the bottom of the pyramid, that leads to the acceptance of higher credit risk at a higher yield that inevitably le...

Full description

Saved in:

Bibliographic Details
Published in:	2022 International Conference on Data Science and Its Applications (ICoDSA) pp. 103 - 108
Main Authors:	Chitty, Rajiv, Gunawikrama, Keerthi, Fernando, Harinda
Format:	Conference Proceeding
Language:	English
Published:	IEEE 06-07-2022
Subjects:	Companies Economics Finance finance companies Industries loan default predication machine learning Machine learning algorithms Sensitivity Support vector machines
Online Access:	Get full text
Tags:	Add Tag No Tags, Be the first to tag this record!

Description
Summary:	Finance Companies (FC's), play a pivotal role in the economy of Sri Lanka, by serving the under banked and non-banked segments of the society. The business model entails lending to the bottom of the pyramid, that leads to the acceptance of higher credit risk at a higher yield that inevitably leads to lower asset quality. The focus on this customer segment has lead to an increase in non performing loans among FCs in the recent past. Due to several challenges facing the industry, including intense competition and lack of experienced credit officers, the FC's have been seeking options to automate evaluation of credit worthiness at the point of loan origination. This work is an attempt to develop a machine learning based loan default prediction system to improve credit decisions. Several traditional machine learning algorithms are chosen, trained and validated by using real world data set related to vehicle leasing, obtained from one of the leading FCs in Sri Lanka. The data set consists of 100,000 cases having 29 attributes each. Models are compared for accuracy, sensitivity, specificity and robustness. The model using Support Vector Machine and Random Forest produces comparatively promising results. Further work is recommended to generalize the model for economic cycles and shocks using micro and macro economic variables.
DOI:	10.1109/ICoDSA55874.2022.9862858