A Knowledge Graph based Approach for Credit Card Fraud Detection

George Konstantinos Dimou ; Georgia Koloniari

Computer Science and Information Systems, Tome 23 (2026) no. 1

Résumé

Credit Card Fraud Detection remains an evolving and difficult challenge due to its complexity, class imbalance, and scale of transactional data. In this work, we propose a novel graph-based approach that constructs a Knowledge Graph (KG) from transactional data to model relationships between different entities. We calculate a variety of centrality measures, both in an unweighted and in a weighted KG, where edge weights represent transaction-specific features like amount, in order to capture structural importance. These centrality measures are then used to enrich the feature space for multiple Machine Learning (ML) models. Our experimental evaluation assesses the performance of the proposed approach with respect to both accuracy and efficiency. Various experiments are conducted, comparing multiple centrality measures, feature combinations and resampling, showcasing how the addition of the centrality measures as classification features significantly improves the performance of our classification models compared to relying only on the original transactional attributes.

Détail

Keywords: Knowledge Graphs, Centrality Measures, Machine Learning, Weighted Graph, Credit Card Fraud Detection