DISCRIMINANT ANALYSIS TO DEVELOP CREDIT SCORING WITH SOCIAL MEDIA DATA: A CASE STUDY IN A SHARIA PEER-TO- PEER LENDING

• Main Author: Andri Putri, Allesandra
• Format: Theses
• Language: Indonesia
• Subjects: Manajemen umum
• Online Access: https://digilib.itb.ac.id/gdl/view/49358
• Institution: Institut Teknologi Bandung

The rapid development of technology has led to the establishment of various financial technology start- ups. One segment of financial technology that is experiencing rapid growth in Indonesia is peer-to-peer lending. PT XYZ, which established in 2017, is one of the sharia-based P2P lending companies in Indonesia. From May to September 2018, the company experienced non-performing loan fluctuations, where the borrowers were unable to pay the principal debts and fees due. Therefore, this study proposes a credit scoring system to determine the credibility of loan applicants while reducing non-performing loans by adding social media data that is currently widely used by internet users. Thus, this study aims to develop a better credit rating model by adding social media variables so that companies can minimize losses and lose opportunities due to misclassification between the borrowers who have the possibility of default and not. This study uses data from 100 samples of borrowers whose loan applications have been approved by PT XYZ. There are five demographic data, two historical payment data, and ten social media data of the borrower used as independent variables to build models with discriminant analysis. Demographic and historical payment data are derived from the company’s data, while social media data collected from the borrower’s Instagram and Facebook accounts. Factor analysis and stepwise methods are used to reduce independent variables to be more efficient and improve the model's accuracy. Based on the results of calculations on training and validation data, stepwise discriminant analysis as a data reduction method has higher predictability than factor analysis, with selected variables namely credit duration, the number of following on Instagram, the borrower’s period of using Instagram, gender, and the number of posts on Facebook at midnight. The use of social media data is also proven to improve the model's prediction accuracy by 4% compared to models that only use demographic data and historical payment data.