mlsanityData to 14 Oct 2026

When Graph Structure Becomes a Liability: A Critical Re-Evaluation of Graph Neural Networks for Bitcoin Fraud Detection under Temporal Distribution Shift

arXiv 2604.19514 ↗First posted 21 Apr 2026cs.LG cs.AI cs.CRcited by 0

Not on the map. Why?

Abstract (arXiv)

The consensus that GCN, GraphSAGE, GAT, and EvolveGCN outperform feature-only baselines on the Elliptic Bitcoin Dataset is widely cited but has not been rigorously stress-tested under a leakage-free evaluation protocol. We perform a seed-matched inductive-versus-transductive comparison and find that this consensus does not hold. Under a strictly inductive protocol, Random Forest on raw features achieves F1 = 0.821 and outperforms all evaluated GNNs, while GraphSAGE reaches F1 = 0.689 +/- 0.017. A paired controlled experiment reveals a 39.5-point F1 gap attributable to training-time exposure to test-period adjacency. Additionally, edge-shuffle ablations show that randomly wired graphs outperform the real transaction graph, indicating that the dataset's topology can be misleading under temporal distribution shift. Hybrid models combining GNN embeddings with raw features provide only marginal gains and remain substantially below feature-only baselines. We release code, checkpoints, and a strict-inductive protocol to enable reproducible, leakage-free evaluation.

Key facts. No citing papers in the map yet (data through 14 Oct 2026). Builds on 18 of its 36 references that are in the map. First posted on arXiv 21 Apr 2026.

Builds on

18 of 36 references.

References · 18

  1. 2016Semi-Supervised Classification with Graph Convolutional Networkscited by 11,361
  2. 2017Graph Attention Networkscited by 8,923
  3. 2015Batch Normalization: Accelerating Deep Network Training by Reducing Internal Covariate Shiftcited by 7,005
  4. 2017Inductive Representation Learning on Large Graphscited by 6,320
  5. 2016XGBoost: A Scalable Tree Boosting Systemcited by 5,166
  6. 2017On Calibration of Modern Neural Networkscited by 4,960
  7. 2017Neural Message Passing for Quantum Chemistrycited by 3,487
  8. 2020Learning under Concept Drift: A Reviewcited by 477
  9. 2020Inductive Representation Learning on Temporal Graphscited by 462
  10. 2019EvolveGCN: Evolving Graph Convolutional Networks for Dynamic Graphscited by 447
  11. 2018Adversarial Attacks on Neural Networks for Graph Datacited by 397
  12. 2014Graph-based Anomaly Detection and Description: A Surveycited by 183
6 more
  1. 2019Anti-Money Laundering in Bitcoin: Experimenting with Graph Convolutional Networks for Financial Forensicscited by 164
  2. 2020A Semi-supervised Graph Attentive Network for Financial Fraud Detectioncited by 151
  3. 2021GraphSMOTE: Imbalanced Node Classification on Graphs with Graph Neural Networkscited by 147
  4. 2022ROLAND: Graph Learning Framework for Dynamic Graphscited by 132
  5. 2018GeniePath: Graph Neural Networks with Adaptive Receptive Pathscited by 66
  6. 2022Inspection-L: Self-Supervised GNN Node Embeddings for Money Laundering Detection in Bitcoincited by 13

Built on by

Not cited yet.