Paper: arXiv 2507.01980
Authors: Linh Nguyen, Marcel Boersma, Erman Acar
Abstract
Fraudulent activity in the financial industry costs billions annually. Detecting fraud, therefore, is an essential yet technically challenging task that requires carefully analyzing large volumes of data. While machine learning (ML) approaches seem like a viable solution, applying them successfully is not so easy due to two main challenges: (1) the sparsely labeled data, which makes the training of such approaches challenging (with inherent labeling costs), and (2) lack of explainability for the flagged items posed by the opacity of ML models, that is often required by business regulations. This article proposes SAGE-FIN, a semi-supervised graph neural network (GNN) based approach with Granger causal explanations for Financial Interaction Networks. SAGE-FIN learns to flag fraudulent items based on weakly labeled (or unlabelled) data points. To adhere to regulatory requirements, the flagged items are explained by highlighting related items in the network using Granger causality. We empirically validate the favorable performance of SAGE-FIN on a real-world dataset, Bipartite Edge-And-Node Attributed financial network (Elliptic++), with Granger-causal explanations for the identified fraudulent items without any prior assumption on the network structure.
Complexity vs Empirical Score
- Math Complexity: 7.5/10
- Empirical Rigor: 7.0/10
- Quadrant: Holy Grail — high math complexity, high empirical rigor
Why this score: The paper employs advanced GNN architectures with formal message-passing equations and Granger-causality for explainability, indicating high math complexity. It is backed by empirical validation on a real-world financial dataset (Elliptic++) and addresses implementation-heavy challenges like semi-supervised learning on bipartite graphs.
Research Flowchart
flowchart TD
A["Research Goal<br>Detecting Fraud in Financial Networks<br>with Low Labels & High Explainability"] --> B["Data Input<br>Elliptic++ Dataset<br>Bipartite Financial Network"]
B --> C["Methodology<br>Semi-Supervised GNN"]
subgraph C ["SAGE-FIN Model"]
direction LR
C1["Graph Structure Learning"] --> C2["Feature Extraction"]
C2 --> C3["Label Propagation"]
end
C --> D["Computational Process<br>Granger Causality Analysis"]
D --> E["Explainable Output<br>Causal Links to Fraudulent Nodes"]
E --> F{"Outcomes"}
F --> G["High Fraud Detection Accuracy"]
F --> H["Regulatory Compliance<br>via Causal Explanations"]
Evidence Card
What the authors claim, on what data, and who has checked it. Read from the paper's own text by a language model: every quote is verbatim, nothing was re-run. How to read this card
Proposes
The paper proposes SAGE-FIN, a semi-supervised graph neural network with Granger-causal explanations for fraud detection in financial interaction networks.
Source sentence
This article proposes SAGE-FIN, a semi-supervised graph neural network (GNN) based approach with Granger causal explanations for Financial Interaction Networks.
Assumes 6
- Labels of financial transactions are expensive to obtain for model training. data
Source sentence
non-fraud) are often expensive to obtain for model training
- The model can be trained on weakly labeled or unlabeled data points. data
Source sentence
SAGE-FIN learns to flag fraudulent items based on weakly labeled (or unlabelled) data points
- No prior assumption on the network structure is required. model
Source sentence
without any prior assumption on the network structure
- The financial transaction data can be represented as a bipartite graph of wallets and transactions. data
Source sentence
The dataset is presented in a bipartite format, where monetary flows from wallets to specific transactions are presented as edges
Show 2 more
- The Elliptic++ dataset contains no missing values. data
Source sentence
Elliptic++ dataset is set up with high quality, ensuring no missing values within its comprehensive set of features
- Granger causality provides a valid causal explanation for flagged fraud items. statistical
Source sentence
transparent causal insights without relying on a specific model structure
Data
- Elliptic++ crypto-onchain public
Source sentence
The real-world Elliptic++ Bitcoin data is utilized. The dataset is presented in a bipartite format, where monetary flows from wallets to specific transactions are presented as edges
Kaggle
Costs
Not applicable
Finds 2
- SAGE-FIN significantly outperformed the Logistic Regression model but did not surpass Random Forest and XGBoost, which had higher precision. performance
Source sentence
Across several model choices, the model significantly outperformed the Logistic Regression model but did not surpass RF and XGB, primarily due to the latter model’s noticeably higher precision
- The SAGE-FIN model demonstrated commendable precision, recall, and F1‑score metrics for anomaly detection on the Elliptic++ dataset. performance
Source sentence
The SAGE-FIN model has demonstrated commendable performance in terms of precision, recall, and F1-score metrics when applied…
Checked by others
No other paper in the archive has been found to support or contradict these findings (compared with 22 related papers). Coverage is partial, so this is neither replication nor refutation.
Code
No repository by the authors found in the paper.
Extracted 2026-11-14 · same card as JSON · A language model chose these items: check the source sentence before relying on one.