Prediction of Loan Defaulters in Micro Finance Using Social Network Data

Faced with growing competition in the micro financing market and higher operational risk, it is ever more important for an MFI to be able to leverage less conventional customer data to improve the efficiency of their lending models. Most MFIs are active in developing countries where fonancial history is generally non existent on their user base which increases the difficulty in assessing the credit worthiness of individuals. Instead, an alternative source of data such as mobile phone call and SMS logs can be utilised to assist with this problem. In this study, call and SMS logs from the borrowers of a MFI operating in the Kenyan marketplace are featurised and used to train various classification models. The results show how such data is a valuable commodity in predicting the default class, particularly when relationship tie-strength features are introduced. The inuence of an existing borrower's loan outcome on a new loan applicant within their social network is also modelled using the spreading activation method as an alternative approach to traditional classification, but results indicate that they are not effective