Year: 2025
Type: article
Abstract: While digital financial transactions have become more convenient, card fraud threats have also risen — especially in card-not-present (CNP) and identity theft cases. The conventional fraud detection mechanisms are unable to tackle the shifting nature of fraud over a period of time, data impairment, and limitation in transparency. We investigate the possibility of applying Large Language Models (LLMs), like GPT, to predictive models of fraud built on structured transaction data that is represented as unstructured natural language for both enhanced detection capability and interpretability. With a combination of LLMs and models such as Logistic Regression and XGBoost, the hybrid system provides higher detection accuracy with human-readable explanations. The paper shows that the generalizing and adaptive nature of LLMs enables them to improve fraud detection systems that comply with regulatory requirements. Keywords: Card Fraud Detection, Large Language Models (LLMs), Explainable Artificial Intelligence (XAI), SMOTE-Tomek, Natural Language Processing (NLP), XGBoost, Logistic Regression.
DOI: 10.55041/ijsrem50999
Authors: Assistant Professor2, Artificial Intelligence & Data Science, GJUS&T HISAR, SUMIT SUMIT
Language: English
Subfield: Artificial Intelligence
Field: Computer Science
Domain: Physical Sciences
Open Access status: bronze
OpenAlex ID: W4411610706