ARTIFICIAL NEURAL NETWORK TO DEVELOP LOAN DEFAULT PREDICTING MODEL WITH SOCIAL MEDIA DATA: A CASE STUDY OF ONLINE PEER TO PEER LENDING

• Main Author: Abdullah Hamzah, Muhammad
• Format: Final Project
• Language: Indonesia
• Online Access: https://digilib.itb.ac.id/gdl/view/64511
• Institution: Institut Teknologi Bandung

Financial technology or fintech is currently developing in the world, especially in Indonesia. One of the fintech that is now also developing in Indonesia is a peer to peer online lending platform. However, one of the main problems is credit risk. There is an online peer to peer lending company in Indonesia that faced an increase of non-performing loan rate. It fluctuates from May to September 2018. Then, the company is trying to develop a credit scoring that adds social media data. So, in this research, it is determined how the influence of social media data on the credit scoring predictability rate and what variables can be used in forming an effective and efficient credit scoring model. In the previous study, the prediction of repayment by borrowers can be improved by social media data. Then in this study, ANN will be used as a loan prediction method. This study used two variables consisting of independent and dependent variables. The dependent variable used is the probability of default. Furthermore, the independent variable consists of the duration of Instagram usage, the frequency of posts in the morning, the frequency of posts in the afternoon, the frequency of posts in the evening, the frequency of posts at night, the frequency of posts in the middle of the night, the number of religious accounts followed on Instagram, followers, following, posts per month on Instagram, the number of posts on Instagram, the tenor, installments, gender, marital status, district, type of work, and the income per month. In this study, it was found that social media data can increase the predictability ratio of 15.8%. Furthermore, five variables can be used as predictors in an effective and efficient credit scoring model because it has the highest influence in predicting default. These variables are the number of follows, the district of residence, the number of posts on Instagram, the frequency of posts at evening, and the tenor. Then, the model that only use these five variables has 4.4% predictability rate higher than the model that only use demographic and historical payment data. Finally, the company is recommended to use social media data in their credit scoring model by paying special attention to the five variables that a high influence in predicting default consisting of the following number, residential district, the number of posts on Instagram, frequency of posts at evening, and tenor. Then also for regulators, supervisors, and other financial institutions can consider the use of social media data to be applied in credit scoring.