Evaluating the Effect of People's Social Network on the Credit Score of Banks and Credit Institutions with Deep Machine Learning and Gradient Methods

This study seeks to investigate the effect of variables and data related to people's social network on their credit score. In this study, two main goals are pursued: reducing information asymmetry and increasing financial inclusion. Achieving the above goals is done by finding meaningful information about people's social data to measure how such data affects their credit score. The basic hypothesis of this study is that people with a high credit score have social relationships with people who are similar and of the same age. A data set of more than 300,000 loans that have been paid by an Iranian bank to real people has been used to confirm and explain the effect of social network variables on credit score. In order to determine the variables of the study, an in-depth interview was conducted with a number of banking experts and people actively involved in the field of credit scoring, and at the end, the variables were determined and classified into three categories: financial, behavioral, and social. The study continued with the logistic regression method and finally with various regression models based on deep machine learning, including gradient. The results of the analyses conducted using the logistic regression method show that, statistically, people's social variables can predict the probability of their loan default. The results of machine learning algorithms also indicate that social network information can significantly improve the performance of loan default prediction.