Graph Neural Networks for predicting loan defaults : a comparative study with traditional ML models

With a newly emerging financial online presence, such as Peer-to-Peer (P2P) lending and a growing loan market for Small and Medium-sizes enterprises, new Machine learning (ML) technologies can be used to predict loan defaults in financial markets. Graph Neural Networks (GNNs) can offer a solution to assess risk for a bank when a company wants to apply for a loan. This research informs about different GNN methods used in recent literature, then chooses GCN as our model to perform research on and establish a framework for developing a GCN in our methodology. An experiment is performed to determine if GCN is indeed better at predicting loan default between loans that have a similar bank. The results show that GCN are performing better in all the benchmarking metrics: recall, precision, AUC (Area under the Curve), and accuracy, compared to Logistic Regression (LR).