Leakage Safe Graph Features for Interpretable Fraud Detection in Temporal Transaction Networks
arXiv:2603.06632v1 Announce Type: new Abstract: Illicit transaction detection is often driven by transaction level attributes however, fraudulent behavior may also manifest through network structure such as central hubs, high flow intermediaries, and coordinated neighborhoods. This paper presents a time respecting, leakage safe (causal) graph feature extraction protocol for temporal transaction networks and evaluates its utility for illicit entity classification. Using the Elliptic dataset, we construct directed transaction graphs and compute interpretable structural descriptors, including degree statistics, PageRank, HITS hub or authority scores, k-core indices, and neighborhood reachability measures. To prevent look ahead bias, we additionally compute causal variants of graph features using only edges observed up to each timestep. A Random Forest classifier trained with strict temporal splits achieves strong discrimination on a held out future test period (ROC-AUC about 0.85, Averag
arXiv:2603.06632v1 Announce Type: new Abstract: Illicit transaction detection is often driven by transaction level attributes however, fraudulent behavior may also manifest through network structure such as central hubs, high flow intermediaries, and coordinated neighborhoods. This paper presents a time respecting, leakage safe (causal) graph feature extraction protocol for temporal transaction networks and evaluates its utility for illicit entity classification. Using the Elliptic dataset, we construct directed transaction graphs and compute interpretable structural descriptors, including degree statistics, PageRank, HITS hub or authority scores, k-core indices, and neighborhood reachability measures. To prevent look ahead bias, we additionally compute causal variants of graph features using only edges observed up to each timestep. A Random Forest classifier trained with strict temporal splits achieves strong discrimination on a held out future test period (ROC-AUC about 0.85, Average Precision about 0.54). Although transaction attributes remain the dominant predictive signal, graph derived features provide complementary interpretability and enable risk context analysis for investigation workflows. We further assess operational utility using Precision at k and evaluate probability reliability via calibration curves and Brier scores, showing that calibrated models yield better aligned probabilities for triage. Overall, the results support causal graph feature extraction as a practical and interpretable augmentation for temporal fraud detection pipelines.
Executive Summary
This article presents a novel framework for detecting illicit transactions in temporal networks by leveraging leakage-safe graph features. The authors employ a time-respecting graph feature extraction protocol that computes interpretable structural descriptors, such as degree statistics and PageRank scores, while preventing look-ahead bias. The proposed method is evaluated on the Elliptic dataset, achieving strong discrimination with a Random Forest classifier and providing complementary interpretability to transaction attributes. The results demonstrate the operational utility of causal graph feature extraction in temporal fraud detection pipelines, with implications for investigation workflows and risk context analysis. The study's findings highlight the potential of graph-based approaches in enhancing the effectiveness and efficiency of illicit transaction detection systems.
Key Points
- ▸ The authors propose a time-respecting, leakage-safe graph feature extraction protocol for temporal transaction networks.
- ▸ The method computes interpretable structural descriptors, including degree statistics and PageRank scores, while preventing look-ahead bias.
- ▸ The proposed approach achieves strong discrimination with a Random Forest classifier on the Elliptic dataset.
Merits
Interpretability and Complementarity
The proposed method provides interpretable structural descriptors that complement transaction attributes, enabling risk context analysis and investigation workflows.
Operational Utility
The study demonstrates the operational utility of causal graph feature extraction in temporal fraud detection pipelines, enhancing the effectiveness and efficiency of illicit transaction detection systems.
Demerits
Limited Dataset
The study is limited to a single dataset (Elliptic), which may not be representative of other temporal networks and fraud detection scenarios.
Dependence on Transaction Attributes
The proposed approach relies heavily on transaction attributes, which may dominate the predictive signal and limit the interpretability of graph-derived features.
Expert Commentary
The article presents a timely and relevant contribution to the field of temporal network analysis, leveraging graph-based approaches to detect illicit transactions. While the study's findings are promising, further research is needed to address the limitations identified, such as the dependence on transaction attributes and the limited dataset. The proposed method has significant implications for the development of effective and efficient illicit transaction detection systems, which can be integrated into existing AML and KYC frameworks. The study's findings also highlight the need for more effective and targeted measures to combat financial crime, informing policy decisions related to the regulation of financial systems.
Recommendations
- ✓ Future studies should investigate the application of the proposed method to other temporal networks and fraud detection scenarios.
- ✓ Researchers should explore the development of more robust and interpretable graph-derived features that can complement transaction attributes and enhance the effectiveness of illicit transaction detection systems.