Enhancing Fraud Transaction Detection via Unlabeled Suspicious Records (GIANTESS Framework)
The paper presents GIANTESS, a novel semi‑supervised fraud detection framework that leverages online‑identified suspicious transactions to augment the feature space, generating pseudo‑labels for out‑of‑distribution samples and employing a hybrid loss to improve detection of covert fraudulent activities, achieving notable recall gains on real‑world datasets.
In the context of rapid digital transformation driven by 5G, AI, and VR, digital trade platforms have proliferated, bringing diverse fraud risks. Ant Financial’s Security TianSuan Lab, in collaboration with Tsinghua University, introduced a new framework called GIANTESS to enhance fraud transaction detection using unlabeled suspicious records.
The research was accepted at the IEEE/ACM International Workshop on Quality of Service (IWQoS) 2024 and received the Best Paper Award.
Traditional fraud detection methods fall into two categories: rule‑based approaches, which rely on expert‑crafted patterns and offer good interpretability but struggle with complex patterns; and machine‑learning approaches, which can capture intricate patterns from large datasets but depend heavily on the quality of the feature space.
Covert fraudulent transactions, which closely resemble benign ones, remain a major challenge because they can evade existing detection systems.
GIANTESS addresses this by:
Generating pseudo‑labels for out‑of‑distribution suspicious samples, thereby expanding the feature space for downstream model training.
Employing a hybrid loss function that jointly leverages real hard labels and pseudo soft labels, focusing the model on transactions that are likely to be mis‑classified as fraud.
The framework operates in two stages: (1) pseudo‑label generation using a universal data‑augmentation‑based training method that smooths the decision boundary of a base classifier, and (2) mixed‑label training that integrates both labeled and suspicious transactions.
Experimental results show that, under a false‑positive rate of 0.1%, GIANTESS improves recall for covert fraud detection in account takeover and scam scenarios by 7.2% and 7.7% respectively, with 45.9% of the newly recalled frauds resembling benign transaction patterns.
The paper includes visual illustrations of the method concept, the GIANTESS workflow, and experimental performance charts.
AntTech
Technology is the core driver of Ant's future creation.
How this landed with the community
Was this worth your time?
0 Comments
Thoughtful readers leave field notes, pushback, and hard-won operational detail here.