Real-Time Behavioural Fraud Prevention in Financial Institutions Using Adaptive Machine Learning

Authors

  • Farainesu Hove Information Technology Department, School of Information Science and Technology, Harare Institute of Technology, Harare, Zimbabwe https://orcid.org/0009-0007-8192-7949
  • Mainford Mutandavari Department of Data Science and Business Systems, SRM Institute of Science and Technology, Kattankulathar, India https://orcid.org/0000-0003-3342-5710

DOI:

https://doi.org/10.38124/ijsrmt.v5i10.1747

Keywords:

Adaptive Machine Learning, Fraud Detection, CRISP-DM, Real-Time Prevention, Decision Tree, Random Forest, XGBoost, Concept Drift, Behavioural

Abstract

The proliferation of digital financial channels has exponentially increased the attack surface for financial fraud, with global card fraud losses reaching USD 33.45 billion in 2022. Conventional rule-based fraud detection systems are inherently incapable of adapting to fast-changing fraud tactics, leading to high levels of undetected fraud and too many false positives impacting customer experience. This paper proposes a real-time behavioural fraud prevention framework based on Adaptive Machine Learning (AML) and CRISP-DM methodology. In this study, three classification algorithms, Decision Tree (DT), Random Forest (RF) and Extreme Gradient Boosting (XGBoost) are trained, evaluated and compared using an anonymised dataset of 499,999 financial transactions extracted from a Zimbabwean commercial bank on imbalanced and class-balanced datasets. We engineer novel behavioural features from raw transactional attributes, namely errorBalanceOrg, errorBalanceDest and hourOfDay that account for 71% of the feature importance of the Random Forest model. The Decision Tree classifier gives the best Area Under the Curve (AUC) of 0.9633361 on the balanced data set, which is a significant improvement over the estimated rule-based baselines. Current adaptive ML trends including federated learning, graph neural networks, transformer based anomaly detection, and concept drift adaptation are also discussed in this paper. In addition, a cloud-native deployment architecture on Microsoft Azure is proposed. The results demonstrate the viability of adaptive machine learning as a high-performance alternative to static rule-based fraud detection in developing-economy banking environments.

Downloads

Download data is not yet available.

Downloads

Published

2026-11-18

How to Cite

Hove, F., & Mutandavari, M. (2026). Real-Time Behavioural Fraud Prevention in Financial Institutions Using Adaptive Machine Learning. International Journal of Scientific Research and Modern Technology, 5(10), 175–184. https://doi.org/10.38124/ijsrmt.v5i10.1747

PlumX Metrics takes 2–4 working days to display the details. As the paper receives citations, PlumX Metrics will update accordingly.

Similar Articles

<< < 35 36 37 38 39 40 41 > >> 

You may also start an advanced similarity search for this article.