Healthcare in Fraudster's Crosshairs: Designing, Implementing and Evaluating a Machine Learning Approach for Anomaly Detection on Medical Prescription Claim Data.
Tizian Matschak, Christoph Prinz, Kristin Masuch, Simon Thanh-Nam Trang · Journal of the Association for Information Systems · 2021
Health insurance claim fraud is a serious problem for the health care industry. As it drives up costs and inefficiency, claim fraud must be effectively detected to provide economic and high-quality healthcare. In this paper, we present a hybrid machine learning approach that was developed, implemented, and evaluated in the context of a prescription claim service provider with data of over one million prescription claims. It combines the advantages of supervised and unsupervised anomaly detection. Based on prescription claim data as well as metadata of providers and patients, we design a data preprocessing pipeline and combine a Random Forest with an Autoencoder. The resulting hybrid model is evaluated against previous research and standard supervised and unsupervised algorithms. Our model outperforms these baseline models with an AUC of 0.8253 and provides an applicable method for medical prescription fraud detection