Abstract
Credit card fraud continues to impose substantial financial losses on banks, merchants, and consumers, and the tactics used by fraudsters shift continuously as detection systems improve. Most existing machine learning approaches to fraud detection are developed and evaluated on static, fixed-period datasets, implicitly assuming that the statistical patterns separating fraudulent and legitimate transactions remain stable over time. In practice, this assumption rarely holds: spending behavior, merchant categories, and fraud strategies evolve, causing a phenomenon known as concept drift, in which a model’s predictive accuracy degrades as the underlying data distribution changes. This study proposes a machine learning framework for detecting credit card fraud that explicitly accounts for this evolving nature of fraudulent behavior, rather than treating fraud detection as a one-time classification problem. The research evaluates and compares several supervised machine learning algorithms, including Logistic Regression, Random Forest, and XGBoost, alongside drift-aware techniques such as sliding- window retraining and drift detection methods, to assess how model performance changes as transaction patterns shift over time. The study addresses the severe class imbalance inherent in fraud datasets through resampling techniques such as SMOTE, and evaluates models using precision, recall, F1-score, and Area Under the Precision-Recall Curve rather than accuracy alone, given the low base rate of fraudulent transactions. By comparing static, one-time-trained models against models that adapt to newly available data, the study aims to determine whether drift-aware approaches offer meaningful improvements in sustained detection performance over time. The findings are expected to provide practical insight into how financial institutions can design fraud detection systems that remain effective as fraud patterns continue to change, rather than relying on models that degrade silently after deployment.
Publication Date
8-2026
Document Type
Thesis
Student Type
Graduate
Degree Name
Professional Studies (MS)
Department, Program, or Center
Graduate Programs & Research
Advisor
Ioannis Karamitsos
Recommended Citation
Alhammadi, Sultan, "Outsmarting Fraud: A Machine Learning Framework for Detecting Evolving Credit Card Fraud Patterns" (2026). Thesis. Rochester Institute of Technology. Accessed from
https://repository.rit.edu/theses/12834
Campus
RIT Dubai
