A Hybrid Machine Learning Framework for Cross-Domain Fraud Detection Using Multimodal Data from Healthcare and Financial Systems

Abstract

Fraud detection remains a critical challenge across domains such as healthcare and finance, where fraudulent behaviors often differ in complexity, structure, and data representation. This paper proposes a hybrid machine learning framework that leverages multimodal data integration to improve cross-domain fraud detection capabilities. The approach combines decision trees, neural networks, and ensemble techniques to capture non-linear and latent patterns in both structured data (e.g., transaction records, billing codes) and unstructured data (e.g., clinical notes, textual claims).

By unifying heterogeneous data sources from healthcare and financial systems, the model demonstrates enhanced adaptability to evolving fraud trends and context-specific anomalies. The hybrid architecture supports deeper pattern recognition across domains and enables improved generalization in environments where data variability and concept drift are prevalent.

Experimental results validate the effectiveness of the proposed approach, showing significant improvements in detection accuracy and robustness compared to baseline single-domain models. This research highlights the potential of cross-domain machine learning and multimodal integration in developing next-generation fraud detection systems.