Inductive Graph Representation Learning for fraud detection

作者：

Highlights：

• Inductive graph embeddings have predictive value for fraud detection.

• Embeddings contain novel information, not captured by transaction features.

• Undersampling enhances predictive performance in highly imbalanced networks.

• GraphSAGE’s inference speed makes it suitable for fraud detection in practice.

• FI-GRL can induce embeddings for many nodes with acceptable delay.

摘要

•Inductive graph embeddings have predictive value for fraud detection.•Embeddings contain novel information, not captured by transaction features.•Undersampling enhances predictive performance in highly imbalanced networks.•GraphSAGE’s inference speed makes it suitable for fraud detection in practice.•FI-GRL can induce embeddings for many nodes with acceptable delay.

论文关键词：Inductive Representation Learning,Graph embeddings,Class imbalance,Fraud detection,GraphSAGE

论文评审过程：Received 9 June 2020, Revised 10 December 2021, Accepted 23 December 2021, Available online 10 January 2022, Version of Record 18 January 2022.

论文官网地址：https://doi.org/10.1016/j.eswa.2021.116463