Reza, Jafrin and Mohammad, Abdul Rahman and Khatun, Mst Murshida and Akter, Fahmida (2026) Human-Centered Explainable AI for Improving Cybersecurity Fraud Investigation and Decision-Making. American Journal of Technology Advancement, 3 (8). ISSN 2997-9382

Text AJTA_Human-Centered+Explainable.pdf Download (1MB)

Abstract

Digital payment systems have come a long way, resulting in a challenge for cybersecurity experts and financial institutions to identify and trace fraudulent credit card transactions. While traditional machine-learning fraud detection systems can deliver high predictive accuracy, they tend to be black-box systems which offer little insight into the reasons behind a fraudulent transaction being classified as such. Lack of transparency can decrease investigator confidence, make the case more difficult to evaluate, and affect the utility of AI in fraud decision making. This study suggests a Human-Centered Explainable Artificial Intelligence (XAI) framework to enhance the process of cybersecurity fraud investigation and decision-making by leveraging a U.S.-based credit-card transaction dataset including transaction, merchant, geographic, temporal and customer-related attributes. The proposed framework combines machine-learning-based classification with explainable AI (XAI) techniques to translate fraud predictions into explainable and actionable evidence in investigations. For comparative fraud classification, Logistic Regression, Random Forest, XGBoost, and LightGBM are considered, and the performance of the models is measured by the precision, recall, F1-score, ROC-AUC, PR-AUC, confusion matrix, false positive rate and false negative rate. In the study, instead of focusing on accuracy only, class imbalance and cost-sensitive learning methods are discussed as they are appropriate. The study does not only focus on accuracy, but on appropriate class-imbalance and cost-sensitive learning methods, since fraudulent transactions make up a minority class. Incorporation of SHAP and LIME to understand the factors that affect fraud predictions, both globally and per transaction. Explanations generated are presented in a way that works for investigators by grouping them into categories of evidence: transaction amount, merchant category, temporal patterns, geographic patterns and customer related attributes, which allows for risk-based interpretation of each alert. The framework also provides a human centered decision layer that translates model outputs and model explanations into risk levels for fraud and priorities for investigations. The proposed approach aims to overcome the drawbacks of opaque fraud detection systems by leveraging interpretable evidence and human judgment to enhance transparency, consistency, and clarity in cybersecurity decision-making. This study offers a practical approach to combining predictive accuracy, explainability, and human oversight in credit-card fraud investigations and lays the groundwork for building reliable AI-powered fraud decision support systems in the financial contexts of the United States.

Item Type:	Article
Uncontrolled Keywords:	Explainable Artificial Intelligence, Cybersecurity Fraud Detection, Human-Centered AI, Credit Card Fraud, Fraud Investigation and Decision Support
Subjects:	H Social Sciences
Depositing User:	admin eprints
Date Deposited:	29 Sep 2026 04:10
Last Modified:	29 Sep 2026 04:10
URI:	http://eprints.umsida.ac.id/id/eprint/17120

Actions (login required)

View Item