Title: Credit scoring with dynamic multilayer graph neural networks
Authors: Maria Oskarsdottir - Reykjavík University (Iceland) [presenting]
Cristian Bravo - Western University (Canada)
Christophe Mues - University of Southampton (United Kingdom)
Kamesh Korangi - University of Southampton (United Kingdom)
Sahab Zandi - Western University (Canada)
Abstract: Credit scoring is one of the oldest applications of data analytics, where lenders use credit scores to help adjudge the risk involved in granting a loan. For most people, access to credit is necessary to support financial wellness, and an acceptable credit score is also required for access to several financial services. While the presence of default correlation has been suspected for a long time, it has only recently been studied to some extent, with the help of network science. Borrowers, in particular, can be connected in different ways, and represented with multilayer networks to reflect various kinds of connections. We present a model for credit risk assessment leveraging a dynamic multilayer network built from a Graph Neural Network and a Recurrent Neural Network. We test our methodology in agricultural lending where sources of connections are geographic location and economic activity of borrowers. The proposed model considers different types of connections between the borrowers as well as the evolution of these connections over time. Preliminary results demonstrate that, when it comes to predicting probability of default of the borrowers, our proposed model brings both better results and interesting insights compared to traditional methods.