Relational Graph Neural Networks for Fraud Detection in a Super-App\n environment

Large digital platforms create environments where different types of user\ninteractions are captured, these relationships offer a novel source of\ninformation for fraud detection problems. In this paper we propose a framework\nof relational graph convolutional networks methods for fraudulent behaviour\nprevention in the financial services of a Super-App. To this end, we apply the\nframework on different heterogeneous graphs of users, devices, and credit\ncards; and finally use an interpretability algorithm for graph neural networks\nto determine the most important relations to the classification task of the\nusers. Our results show that there is an added value when considering models\nthat take advantage of the alternative data of the Super-App and the\ninteractions found in their high connectivity, further proofing how they can\nleverage that into better decisions and fraud detection strategies.\n