Work

Understanding Structured Financial Data with LLMs: A Case Study on Fraud Detection

Year: 2026

Type: conference-paper

Abstract: Detecting fraud in financial transactions typically relies on tabular models that demand heavy feature engineering to handle highdimensional data and offer limited interpretability, making it difficult for humans to understand predictions.Large Language Models (LLMs), in contrast, can produce human-readable explanations and facilitate feature analysis, potentially reducing the manual workload of fraud analysts and informing system refinements.However, they perform poorly when applied directly to tabular fraud detection due to the difficulty of reasoning over many features, the extreme class imbalance, and the absence of contextual information.To bridge this gap, we introduce FinFRE-RAG, a two-stage approach that applies importance-guided feature reduction to serialize a compact subset of numeric/categorical attributes into natural language and performs retrieval-augmented in-context learning over label-aware, instance-level exemplars.Across four public fraud datasets and three families of open-weight LLMs, FinFRE-RAG substantially improves F1/MCC over direct prompting and is competitive with strong tabular baselines in several settings.Although these LLMs still lag behind specialized classifiers, they narrow the performance gap and provide interpretable rationales, highlighting their value as assistive tools in fraud analysis.* Work done during internship at Coinbase.This paper contains the author's personal opinions and does not constitute a company policy or statement.These opinions are not endorsed by or affiliated with Coinbase, Inc. or its subsidiaries.

Source: Proceedings of the conference - Association for Computational Linguistics. Meeting

DOI: 10.18653/v1/2026.acl-long.1071

Authors: Xuwei Tan, Yao Ma, Xueru Zhang

Institution: The Ohio State University

Language: English

Topic: Imbalanced Data Classification Techniques

Subfield: Artificial Intelligence

Field: Computer Science

Domain: Physical Sciences

Open Access status: hybrid

Funders: National Science Foundation, Division of Civil, Mechanical and Manufacturing Innovation

Awards: Collaborative Research: FW-HTF-R: Future of Construction Workplace Health Monitoring, Collaborative Research: RI: AF: Small: Long-Term Impact of Fair Machine Learning under Strategic Individual Behavior, Collaborative Research: RI: AF: Small: Long-Term Impact of Fair Machine Learning under Strategic Individual Behavior, MATH-DT: Heterogeneous transfer and federated learning for digital twin in unmanned aerial vehicles, Collaborative Research: SLES: Long-Term Safety for Human-AI Ecosystem