CleverCatch: A Knowledge-Guided Weak Supervision Model for Fraud Detection

TL;DR

CleverCatch tackles healthcare fraud detection under extreme label scarcity by embedding domain rules into a shared latent space alongside data representations. It learns soft, geometry-based rule representations through Rule and Sample Encoders trained on synthetic compliance/violation signals, with pseudo-labels generated via optimal transport to guide a base anomaly detector. Empirical results on Medicare Part D show improved recall and competitive AUC, with cost-preference rules contributing most to gains and opioid rules providing complementary signals, while ablations confirm the value of knowledge-guided alignment. The approach offers enhanced interpretability and scalability for high-stakes domains and can extend to richer multi-modal signals and real-time monitoring.

Abstract

Healthcare fraud detection remains a critical challenge due to limited availability of labeled data, constantly evolving fraud tactics, and the high dimensionality of medical records. Traditional supervised methods are challenged by extreme label scarcity, while purely unsupervised approaches often fail to capture clinically meaningful anomalies. In this work, we introduce CleverCatch, a knowledge-guided weak supervision model designed to detect fraudulent prescription behaviors with improved accuracy and interpretability. Our approach integrates structured domain expertise into a neural architecture that aligns rules and data samples within a shared embedding space. By training encoders jointly on synthetic data representing both compliance and violation, CleverCatch learns soft rule embeddings that generalize to complex, real-world datasets. This hybrid design enables data-driven learning to be enhanced by domain-informed constraints, bridging the gap between expert heuristics and machine learning. Experiments on the large-scale real-world dataset demonstrate that CleverCatch outperforms four state-of-the-art anomaly detection baselines, yielding average improvements of 1.3\% in AUC and 3.4\% in recall. Our ablation study further highlights the complementary role of expert rules, confirming the adaptability of the framework. The results suggest that embedding expert rules into the learning process not only improves detection accuracy but also increases transparency, offering an interpretable approach for high-stakes domains such as healthcare fraud detection.

Figures (3)

• Figure 1: Prescription behavior and fraud patterns: (a) Total cost vs. total claims for all physician-drug combinations; (b) Mean drug cost per claim by specialty; (c) Average daily supply per claim; (d) Average cost per claim between fraud-labeled and non-fraud physicians.
• Figure 2: (a) Cumulative distribution of drug pair similarities by shared protein targets; (b) Box plot of price differences, grouped by whether fraud or non-fraud physicians more often prescribed the costlier drug; (c) Distribution of price differences, grouped as in (b).
• Figure 3: Workflow of CleverCatch. The system embeds domain rules and prescription data into a shared latent space via rule and sample encoders. Synthetic compliance/violation samples guide alignment, which generates pseudo-labels to enhance the base anomaly detection model.