Year: 2026
Type: article
Abstract: Digital payment systems need to flag potentially fraudulent transactions while keeping alert volumes manageable for analysts. This paper presents a prototype that processes structured transaction data, derives behavioral features, and uses supervised classifiers to assign a continuous fraud risk score to each transaction. Configurable thresholds map scores to three categories: Genuine, Suspicious, and High-Risk. A React dashboard displays transaction scores, prioritizes alerts, and can show the features that contributed to a score. The evaluation design compares logistic regression, random forest, and XGBoost with a rule-based baseline using public credit card data and synthetic transactions. Illustrative scores demonstrate the comparison across classification and ranking metrics. The paper describes the serving architecture, the latency reported for the prototype, and the additional testing required to assess detection performance and latency under operational conditions
Language: English
Subfield: Artificial Intelligence
Field: Computer Science
Domain: Physical Sciences
Open Access status: diamond
OpenAlex ID: W7236220273
