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

What problem does this paper attempt to address?

The main problem that this paper attempts to solve is **how to use Relational Graph Neural Networks (R - GCNs) for fraud detection in the Super - App environment**. Specifically, the authors focus on the potential fraud behavior of newly - registered users when using financial - related services, and hope to improve the accuracy of fraud detection by constructing and comparing different graph neural network models. ### Specific description of the problem 1. **Early identification of fraud behavior**: In particular, when new users register and start using financial - related services, how to identify the fraud intentions of these users. Since new users lack sufficient historical data, traditional machine - learning models perform poorly in this case, so new methods need to be explored. 2. **Utilizing multi - source data in the Super - App**: The Super - App contains not only transaction data, but also user behavior data, device information, credit card information, etc. These data can capture the complex relationships between users, thereby providing more information for fraud detection. 3. **Evaluating the effectiveness of different graph neural network architectures**: By constructing different graph neural network models (such as RGCN1, RGCN2, RGCN3, RGCN4), evaluate their performance in the fraud - detection task and find the model architecture that is most suitable for the Super - App environment. 4. **Interpreting the prediction results of the model**: By using interpretability algorithms such as GNNExplainer, determine which relationships are most important for fraud detection, thereby helping to understand the decision - making process of the model. ### Main contributions of the paper - Proposed a framework based on relational graph neural networks for detecting the fraud behavior of new users in the Super - App environment. - Verified the effects of different graph neural network architectures through experiments, and found that the RGCN3 model, which combines user historical data and non - financial user characteristics, performs best. - Used the GNNExplainer algorithm to reveal the most critical relationships in the model, such as the relationships between users and devices and credit cards. ### Summary The core objective of this paper is to use the multi - source data in the Super - App and graph neural network technology to develop an effective fraud - detection method, especially for the potential fraud behavior of newly - registered users in financial - related services. By comparing the performance of different models, the authors have proven that the model that combines user historical data and relationship information can significantly improve the effect of fraud detection.

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

Supporting Financial Inclusion with Graph Machine Learning and Super-App Alternative Data

FdGars: Fraudster Detection via Graph Convolutional Networks in Online App Review System

DGA-GNN: Dynamic Grouping Aggregation GNN for Fraud Detection

Fraud Detection Based on Graph Neural Network

Dynamic Fraud Detection: Integrating Reinforcement Learning into Graph Neural Networks

Research on Fraud Detection Method Based on Heterogeneous Graph Representation Learning

A Semi-supervised Graph Attentive Network for Financial Fraud Detection

GraphGuard: Contrastive Self-Supervised Learning for Credit-Card Fraud Detection in Multi-Relational Dynamic Graphs

Enhancing fraud detection in banking by integration of graph databases with machine learning

Feature-Level Fusion of Super-App and Telecommunication Alternative Data Sources for Credit Card Fraud Detection

Dynamic Relation-Attentive Graph Neural Networks for Fraud Detection

Heterogeneous Graph Auto-Encoder for CreditCard Fraud Detection

Towards Consumer Loan Fraud Detection: Graph Neural Networks with Role-Constrained Conditional Random Field

Transaction Fraud Detection via an Adaptive Graph Neural Network

Spatio-Temporal Graph Representation Learning for Fraudster Group Detection

Graph Neural Network for Fraud Detection via Spatial-temporal Attention

2SFGL: A Simple And Robust Protocol For Graph-Based Fraud Detection

Loan Fraud Users Detection in Online Lending Leveraging Multiple Data Views

A Graph-Powered Large-Scale Fraud Detection System

Towards Federated Graph Learning for Collaborative Financial Crimes Detection