Modeling Cardholder Spending Rhythms With Hierarchical Temporal Attention for Early Fraud Detection in Streaming Transactions
INTERNATIONAL JOURNAL OF RESEARCH & SOCIAL SCIENCES · 25.06.2026 · 6 (1) · 194-204
DOI: 10.66238/ijrss.v6i1.106 ↗Kurz gefasst
Fachartikel in der Zeitschrift INTERNATIONAL JOURNAL OF RESEARCH & SOCIAL SCIENCES, online erschienen am 25.06.2026. Verfasst von Admin Admin. Der Volltext ist frei zugänglich (Open Access, CC BY 4.0).
Abstract (Originalsprache)
The proliferation of digital payment platforms has intensified fraudulent activity across global financial networks, compelling institutions to develop detection architectures capable of operating at transaction-stream velocity while preserving high precision. This paper proposes a Hierarchical Temporal Attention (HTA) framework that encodes individual cardholder spending rhythms across intra-day, weekly, and monthly granularities, integrating periodic behavioral embeddings with a hierarchical convolutional encoder-decoder scoring network for real-time fraud detection in streaming transaction environments. The rhythm encoding component exploits the von Mises circular distribution to model the periodic structure of transaction timing, while the encoder-decoder architecture progressively abstracts multi-scale temporal features through symmetric downsampling and upsampling pathways. Experiments on a large-scale real-world credit card dataset demonstrate that HTA achieves an area under the receiver operating characteristic curve (AUC-ROC) of 0.9847, outperforming baseline models including long short-term memory (LSTM), Transformer, and XGBoost under both partial and full ensemble evaluation conditions. The framework exhibits particular strength in detecting first-occurrence fraud events by rapidly identifying deviations from established spending rhythms, validating the core hypothesis that temporal periodicity constitutes a powerful and underexploited fraud signal.
Zitieren
APA
Admin, A. (2026). Modeling Cardholder Spending Rhythms With Hierarchical Temporal Attention for Early Fraud Detection in Streaming Transactions. INTERNATIONAL JOURNAL OF RESEARCH & SOCIAL SCIENCES, 6(1), 194-204. https://doi.org/10.66238/ijrss.v6i1.106
Vancouver
Admin A. Modeling Cardholder Spending Rhythms With Hierarchical Temporal Attention for Early Fraud Detection in Streaming Transactions. INTERNATIONAL JOURNAL OF RESEARCH & SOCIAL SCIENCES. 2026;6(1):194-204. doi:10.66238/ijrss.v6i1.106
BibTeX
@article{Admin2026,
author = {Admin, Admin},
title = {Modeling Cardholder Spending Rhythms With Hierarchical Temporal Attention for Early Fraud Detection in Streaming Transactions},
journal = {INTERNATIONAL JOURNAL OF RESEARCH & SOCIAL SCIENCES},
year = {2026},
volume = {6},
number = {1},
pages = {194--204},
doi = {10.66238/ijrss.v6i1.106}
}
EUVIMED