Research Article

Adversarial Machine Learning for Robust Fraud Detection in High-Frequency Financial Transactions

Authors

  • Mohammad Kowshik Alam Master of science in Business Analytics, Grand Canyon University, Arizona ,USA
  • Md Asief Mahmud Master of Science in Business Analytics, Grand Canyon University, Arizona, USA
  • MD ASHRAFUL ALAM Master of Science in Business Analytics, Trine University, Arizona, USA

Abstract

The fast development of financial technologies has led to the sophistication of fraud and the frequency of fraudulent activities and has become an immense threat to critical financial infrastructure and a deterrent of the trust of people to financial systems. Due to the increasing volumes of transactions and sophistication of fraud strategies, conventional systems of detection based on rules are no longer effective in detecting anomalies arising in real time. This study will focus on the possibilities of finding solutions using artificial intelligence (AI) and machine learning (ML) to identify fraud and anomaly in financial transactions and increase cybersecurity resilience. This paper uses a high-fidelity synthetic fraud detection dataset with 50,000 transactions with 21 various characteristics, such as user profiles, kinds of transactions, risk scores, unique device characteristics, and numerous past indicators of frauds, to establish a powerful analytical approach to the problem. The approach will utilize preprocessing and feature engineering in Python, visual analytics done in Tableau to discover patterns and predictive modeling in XGBoost, a gradient boosting algorithm that can handle imbalanced tasks effectively. Exploratory data analysis indicates a strong imbalance in classes since only approximately 1% of all transactions is fraud due to which more sophisticated modeling methodologies need to be applied. Visually, it is easy to identify all known significant trends in transaction quantities, geographic placements, time-span ratios, and prior records of fraud that can be acted upon to control risky territories and actions. The precision and recall of the model based on XGBoost are high due to the success of the model to identify rare fraudulent transactions and differentiate their occurrence with the authentic cases. These results validate that both AI and ML can model latent fraud patterns in real-time, lower the false positive rate, and produce insightful information that can be understood by legal authorities that will minimize the risks in advance by financial institutions. With the implementation of intelligent detection systems within the financial practices, organizations can maximize their security, the use of their resources, as well as make dynamic responses to the changing threats. That is why, in the study, a lot of emphasis is laid on the necessity of including geo-behavioral and temporal characteristics to ensure that the models perform better and become context friendly. In future, new topics will be real-time deployment, streaming data analysis, factions of robustness against adversaries and ethical compliance of AI-based decision-making. The study offers a flexible, responsive, and smart solution to preventing fraud and cyberattacks on financial ecosystems, which are increasingly posing a threat to financial systems and institutions.

Article information

Journal

Journal of Computer Science and Technology Studies

Volume (Issue)

7 (8)

Pages

314-335

Published

2025-08-03

How to Cite

Mohammad Kowshik Alam, Md Asief Mahmud, & MD ASHRAFUL ALAM. (2025). Adversarial Machine Learning for Robust Fraud Detection in High-Frequency Financial Transactions. Journal of Computer Science and Technology Studies, 7(8), 314-335. https://doi.org/10.32996/jcsts.2025.7.8.35

Downloads

Views

0

Downloads

0

Keywords:

Fraud Detection, Anomaly Detection, Machine Learning, Artificial Intelligence, and Cybersecurity Financial Transactions