Unsupervised Anomaly Detection of Healthcare Providers using Generative Adversarial Network

Abstract

Healthcare fraud is considered a challenge for many societies. Healthcare funding that could be spent on medicine, care for the elderly or emergency room visits is instead lost to fraudulent activities by medical practitioners or patients. With rising healthcare costs, healthcare fraud is a major factor in increasing healthcare costs. This study evaluates previous anomaly detection machine learning models and proposes an unsupervised framework to identify anomalies using a Generative Adversarial Network (GAN) model. The GAN anomaly detection model was applied to two different healthcare provider data sets. The anomalous healthcare providers were further analysed through the application of classification models with the logistic regression and extreme gradient boosting models showing acceptable performances. Results from the SHapley Additive exPlanation also shows the predictors used to explain the anomalous healthcare providers.