Identifying frauds and anomalies in Medicare-B dataset

Abstract

Healthcare industry is growing at a rapid rate to reach a market value of 7trilliondollarsworldwide.Atthesametime,fraudinhealthcareisbecomingaseriousproblem,amountingto5100 billion dollars each year in US. Manually detecting healthcare fraud requires much effort. Recently, machine learning and data mining techniques are applied to automatically detect healthcare frauds. This paper proposes a novel PageRank-based algorithm to detect healthcare frauds and anomalies. We apply the algorithm to Medicare-B dataset, a real-life data with 10 million healthcare insurance claims. The algorithm successfully identifies tens of previously unreported anomalies.