FROM RULES TO GRAPH LEARNING: A REVIEW OF COMPUTATIONAL APPROACHES TO DETECT SUSPICIOUS BANK TRANSACTIONS
Journal of Information System and Technology Management · 25.06.2026 · 11 (43) · 132
DOI: 10.35631/jistm.1143008 ↗Kurz gefasst
Fachartikel in der Zeitschrift Journal of Information System and Technology Management, online erschienen am 25.06.2026. Verfasst von Nordaliela Mohd Rusli und 1 weiteren Autor:innen. Die Arbeit nennt 45 Literaturangaben.
Abstract (Originalsprache)
Automated systems for detecting suspicious transaction were introduced in the early 2000s to manage the growing volume and complexity of financial transactions. Subsequent technological advancements have driven the convergence of Artificial Intelligence (AI) and Big Data Analytics for transaction data analysis. However, these advancements come at considerable implementation cost and despite significant investment, critical limitations persist that require attention to mitigate financial losses. This review maps the evolution of approaches used to detect suspicious bank transactions between 2012 and 2025, and analyses the datasets employed in the reviewed studies. Articles were sourced from Scopus-indexed journals using the search term "money laundering detection", and the collection was restricted to computer science publications focusing on suspicious bank transaction detection. The screening process excluded studies on conceptual profiling, risk assessment, Hawala networks, and fraud. The remaining articles were then analysed to extract identified problems and limitations, proposed solutions, algorithms, and dataset characteristics. Datasets were classified into three categories: forensic criminal investigation datasets, operational bank AML/SAR-level datasets, and synthetic or simulator-based datasets. The findings indicate a clear methodological progression from rule-based and clustering approaches to supervised machine learning, and subsequently to graph-based and deep learning models. However, more than 20% of the reviewed studies relied on synthetic datasets which are unvalidated using actual money laundering activity. Given the absence of detailed forensic insights in such datasets, findings derived from synthetic data must be interpreted with caution. Beyond detection methods and algorithms, dataset realism is equally critical, particularly in the context of public policy and banking practice.
Zitieren
APA
Rusli, N. M., & Zainal, A. (2026). FROM RULES TO GRAPH LEARNING: A REVIEW OF COMPUTATIONAL APPROACHES TO DETECT SUSPICIOUS BANK TRANSACTIONS. Journal of Information System and Technology Management, 11(43), 132. https://doi.org/10.35631/jistm.1143008
Vancouver
Rusli NM, Zainal A. FROM RULES TO GRAPH LEARNING: A REVIEW OF COMPUTATIONAL APPROACHES TO DETECT SUSPICIOUS BANK TRANSACTIONS. Journal of Information System and Technology Management. 2026;11(43):132. doi:10.35631/jistm.1143008
BibTeX
@article{Rusli2026,
author = {Rusli, Nordaliela Mohd and Zainal, Anazida},
title = {FROM RULES TO GRAPH LEARNING: A REVIEW OF COMPUTATIONAL APPROACHES TO DETECT SUSPICIOUS BANK TRANSACTIONS},
journal = {Journal of Information System and Technology Management},
year = {2026},
volume = {11},
number = {43},
pages = {132},
doi = {10.35631/jistm.1143008}
}
EUVIMED